Device parameter determination using latent variables

Abstract

Presented herein are methods and systems for configuring a plurality of operational parameters of a recipient device using latent variables. A plurality of predicted operational parameters of a recipient device (e.g., implantable medical device) is generated, via a decoder of a trained autoencoder network, based on a latent vector having a plurality of latent variables. A difference between the plurality of predicted operational parameters and a plurality of observed operational parameters of the recipient device is iteratively backpropagated to generate an updated latent vector. A plurality of operational parameters is generated based on the updated latent vector, and the recipient device is configured using the plurality of operational parameters.

Description

Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1DEVICE PARAMETER DETERMINATION USING LATENT VARIABLESBACKGROUNDField of the Invention[ooot] The present invention relates generally to configuring one or more operational parameters of a recipient device based on predicted parameters generated using latent variables.Related Art

[0002] Medical devices have provided a wide range of therapeutic benefits to recipients over recent decades. Medical devices can include internal or implantable components / devices, external or wearable components / devices, or combinations thereof (e.g., a device having an external component communicating with an implantable component). Medical devices, such as traditional hearing aids, partially or fully-implantable hearing prostheses (e.g., bone conduction devices, mechanical stimulators, cochlear implants, etc.), pacemakers, defibrillators, functional electrical stimulation devices, and other medical devices have been successful in performing lifesaving and / or lifestyle enhancement functions and / or recipient monitoring for a number of years.

[0003] The types of medical devices and the ranges of functions performed thereby have increased over the years. For example, many medical devices, sometimes referred to as “implantable medical devices,” now often include one or more instruments, apparatus, sensors, processors, controllers or other functional mechanical or electrical components that are permanently or temporarily implanted in a recipient. These functional devices are typically used to diagnose, prevent, monitor, treat, or manage a disease / injury or symptom thereof, or to investigate, replace or modify the anatomy or a physiological process. Many of these functional devices utilize power and / or data received from external devices that are part of, or operate in conjunction with, implantable components.SUMMARY

[0004] In one aspect, a method is provided. The method comprises: generating, via a decoder of a trained autoencoder network, a plurality of predicted operational parameters of an implantable medical device based on a latent vector having a plurality of latent variables; iteratively backpropagating a difference between the plurality of predicted operational parameters and a plurality of observed operational parameters of the implantable medical device to generate an updated latent vector; generating, via the decoder, a plurality ofAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 operational parameters based on the updated latent vector; and configuring the implantable medical device with the plurality of operational parameters.

[0005] In one aspect, a second method is provided. The method comprises: sampling a latent space to obtain one or more latent vectors each having a plurality of latent variables; generating, via a decoder of a trained autoencoder network, one or more vectors each having a plurality of predicted stimulation parameters for a recipient device based the one or more latent vectors; determining a prediction uncertainty of the one or more vectors; and configuring one or more stimulation parameters for the recipient device based on the prediction uncertainty.

[0006] In another aspect, a system is provided. The system comprises: a memory; and at least one processor operable coupled to the memory, wherein the at least one processor is configured to: generate, via a decoder of a trained autoencoder network, a plurality of predicted unobserved stimulation parameters for a recipient device; select one or more electrodes of the recipient device to be configured based on the plurality of predicted unobserved stimulation parameters; and configure one or more unobserved stimulation parameters of the one or more electrodes.

[0007] In another aspect, one or more non-transitory computer readable storage media are provided. The one or more non-transitory computer readable storage media comprising instructions that, when executed by a processor, cause the processor to: generate, via a trained machine learning model, a plurality of operational parameters; determine a prediction uncertainty of the plurality of operational parameters; select one or more electrodes of a recipient device based on the prediction uncertainty; and configure the one or more electrodes based on the plurality of operational parameters.

[0008] In another aspect, a system is provided. The system comprises: a memory; and at least one processor operable coupled to the memory, wherein the at least one processor is configured to: obtain one or more latent vectors each having a plurality of latent variables; generate, via a decoder of a trained autoencoder network, one or more predicted vectors each having a plurality of predicted stimulation parameters for a device based the one or more latent vectors; and configure one or more stimulation parameters for the device based on the plurality of predicted stimulation parameters.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Embodiments of the present invention are described herein in conjunction with the accompanying drawings, in which:

[0010] FIG. 1A is a schematic diagram illustrating a cochlear implant system with which aspects of the techniques presented herein can be implemented;[ooit] FIG. IB is a side view of a recipient wearing a sound processing unit of the cochlear implant system of FIG. 1A;

[0012] FIG. 1C is a schematic view of components of the cochlear implant system of FIG. 1 A;

[0013] FIG. ID is a block diagram of the cochlear implant system of FIG. 1A;

[0014] FIG. 2 is a functional block diagram illustrating a system for generating a plurality of predicted parameters for configuring a recipient device;

[0015] FIG. 3 is a functional block diagram illustrating another system for generating a plurality of predicted parameters for configuring a recipient device;

[0016] FIG. 4 is a functional block diagram illustrating a system for generating, via a trained decoder network, a plurality of predicted parameters for configuring a recipient device;

[0017] FIG. 5 is a functional block diagram illustrating a system for determining uncertainty values associated with a plurality of predicted parameters;

[0018] FIG. 6 is a graphical depiction illustrating a graph for visualizing uncertainty values associated with a plurality of predicted parameters;

[0019] FIG. 7 is a flowchart illustrating a method for configuring an implantable medical device with a plurality of operational parameters;

[0020] FIG. 8 is a flowchart illustrating a method for configuring one or more stimulation parameters for a recipient device based on prediction uncertainty;

[0021] FIG. 9 is a flowchart illustrating operations performed by a system comprising a memory and at least one processor operable coupled to the memory, wherein the at least one processor is configured to perform the operations;

[0022] FIG. 10 is a flowchart illustrating operations for configuring one or more electrodes based on a plurality of operational parameters, wherein the operations are performed by a processor executing instructions stored in one or more non-transitory computer readable storage media;Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0023] FIG. 11 is a flowchart illustrating operations performed by a system comprising a memory and at least one processor operable coupled to the memory, wherein the at least one processor is configured to perform the operations;

[0024] FIG. 12 is a schematic diagram illustrating a vestibular stimulator system with which aspects of the techniques presented herein can be implemented;

[0025] FIG. 13 is a schematic diagram illustrating a retinal prosthesis system with which aspects of the techniques presented herein can be implemented; and

[0026] FIG. 14 illustrates a cochlear implant fitting system with which aspects of the techniques presented herein can be implemented.DETAILED DESCRIPTION

[0027] Presented herein are techniques for configuring one or more operational settings or parameters, such as stimulation settings / parameters (e.g., threshold levels (“T-levels”), comfort levels (“C -levels”), a dynamic range, etc.), of a recipient device (e.g., implantable medical device) using one or more latent variables.

[0028] More specifically, certain implantable medical devices, such as cochlear implants, electro-acoustic hearing prosthesis, auditory brainstem implants, etc., operate by converting at least a portion of received sound signals into electrical stimulation signals for delivery to a recipient’s auditory system. The window / range of electrical charges (controlled via amplitude, duration or both) at which electrical stimulation signals may be delivered to the recipient’s auditory system is limited. In particular, if the charge of the electrical stimulation signals is too low, then the associated sounds used to generate the electrical stimulation signals will not be perceived by the recipient (i.e., the stimulation signals will either not evoke a neural response in the cochlea or evoke a neural response that cannot be perceived by the recipient). Conversely, if the charge of the electrical stimulation signals is too high, then the associated sounds used to generate the electrical stimulation signals will be perceived as too loud or uncomfortable by the recipient.

[0029] As such, electrical stimulation signals are generally delivered between a lower limit, referred to herein as a “threshold level,” at which the associated sound signals are barely audible to the recipient, and an upper limit, referred to herein as a “comfort level,” above which the associated sound signals are uncomfortably loud to the recipient. The difference in electrical stimulations between the threshold level and the comfort level is referred to herein as the “dynamic range.” In general, the term “stimulation parameters” herein can include theAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 threshold level, the comfort level, the dynamic range, or other attributes (e.g., rate, frequency band, modulation, etc.) corresponding to electrical stimulation signals to be delivered to a recipient, regardless of whether or not the electrical stimulation signals are generated based on sound signals.

[0030] In the specific example of cochlear implants, due to a recipient’s specific anatomical features, the insertion depth of a given electrode, or other variables, the dynamic range may be different for different electrodes implanted in a recipient. That is, different electrodes implanted in a recipient may have different associated threshold and comfort levels. The range in acoustic amplitudes of sound signals received by a cochlear implant (or other auditory prosthesis) is considerably larger than the dynamic range associated with an electrode. As such, the conversion of the received sound signals into electrical stimulation signals for delivery to the recipient includes, among other operations, mapping (compression) of the acoustic amplitudes into electrical amplitudes within the dynamic range of the corresponding electrode(s) (i.e., the stimulating contact(s) at which the electrical stimulation is delivered to the recipient).

[0031] Medical practitioners (e.g., clinicians) often participate in clinical sessions to configure a recipient device (e.g., medical device) that is worn by, implanted in, etc., a recipient. These clinical sessions could be performed for a variety of reasons, such as to initially “fit” or configure (e.g., program) the recipient device for the recipient, to adjust or refine parameters / settings of the recipient device, to troubleshoot problems with the recipient device, etc. In conventional clinical sessions, medical practitioners configure or adjust operational parameters of a recipient device based on feedback from the recipient. In the specific example of cochlear implants, medical practitioners (e.g., audiologists) often need to perform new measurements to determine operational parameters (e.g., C-levels or T-levels) during configuration, or “fitting,” sessions. In many cases, due to a lack of information on the operational parameters, a medical practitioner must perform measurements on all electrodes of an electrode array even if some of the electrodes may not require configuration. This process can be time-consuming and inconvenient for the recipient. Thus, it is desirable to leverage data modeling techniques to predict / estimate operational parameters that have not yet been measured.

[0032] However, conventional data modeling techniques can only accommodate fixed-length inputs and outputs. These conventional techniques are not suitable for estimating operational parameters for a recipient device due to dynamic size and data availability of the input and / orAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 output data. For example, the size of the input and / or output data associated with operational parameter prediction changes each time new data (e.g., new measurements) is generated / provided. As such, a new data model must be developed to accommodate the changing input and output data. Thus, conventional data models are impractical as they cannot dynamically generate parameter predictions for variable -length data. Moreover, conventional data modeling techniques can only model fixed-length input data having a specific order, thus medical practitioners lack the flexibility to perform measurements in any sequence they prefer. This further increases inefficiency in the device configuration process.

[0033] In order to address the above and other challenges, the techniques presented herein leverage a machine learning model, such as an autoencoder network, which incorporates latent variables in predicting / estimating operational or stimulation parameters. By generating predictions based on latent variables, the techniques presented herein can dynamically accommodate variable-length and randomly ordered input data (e.g., observed operational parameters). Moreover, the techniques presented herein include determining and providing prediction uncertainty associated with parameters predicted by the machine learning model to assist medical practitioners in deciding which prediction may be reliable, and which parameter requires further configuration. Thus, based on the predicted parameters and / or associated prediction uncertainty information, the medical practitioner can speed up the device configuration process by quickly identifying which electrode requires measurement. Further, the techniques presented herein include iteratively refining predicted parameters to minimize the difference between predicted and observed parameters, thus providing continuous improvement of predictions generated by the machine learning model.

[0034] There are a number of different types of devices in / with which embodiments of the present invention may be implemented. Merely for ease of description, the techniques presented herein are primarily described with reference to a specific device in the form of a cochlear implant system. However, it is to be appreciated that the techniques presented herein may also be partially or fully implemented by any of a number of different types of devices, including consumer electronic device (e.g., mobile phones), wearable devices (e.g., smartwatches), hearing devices, implantable medical devices, consumer electronic devices, etc. As used herein, the term “hearing device” is to be broadly construed as any device that acts on an acoustical perception of an individual, including to improve perception of sound signals, to reduce perception of sound signals, etc. In particular, a hearing device can deliver sound signals to a user in any form, including in the form of acoustical stimulation, mechanicalAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 stimulation, electrical stimulation, etc., and / or can operate to suppress all or some sound signals. As such, a hearing device can be a device for use by a hearing-impaired person (e.g., hearing aids, middle ear auditory prostheses, bone conduction devices, direct acoustic stimulators, electro-acoustic hearing prostheses, auditory brainstem stimulators, bimodal hearing prostheses, bilateral hearing prostheses, dedicated tinnitus therapy devices, tinnitus therapy device systems, combinations or variations thereof, etc.), a device for use by a person with normal hearing (e.g., consumer devices that provide audio streaming, consumer headphones, earphones, and other listening devices), a hearing protection device, etc. In other examples, the techniques presented herein can be implemented by, or used in conjunction with, various implantable medical devices, such as visual devices (i.e., bionic eyes), sensors, pacemakers, drug delivery systems, defibrillators, functional electrical stimulation devices, catheters, seizure devices (e.g., devices for monitoring and / or treating epileptic events), sleep apnea devices, electroporation devices, etc.

[0035] FIGs. 1A-1D illustrates an example cochlear implant system 102 with which aspects of the techniques presented herein can be implemented. The cochlear implant system 102 comprises an external component 104 and an implantable component 112. In the examples of FIGs. 1A-1D, the implantable component is sometimes referred to as a “cochlear implant.” FIG. 1A illustrates the cochlear implant 112 implanted in the head 154 of a user, while FIG. IB is a schematic drawing of the external component 104 worn on the head 154 of the user. FIG. 1C is another schematic view of the cochlear implant system 102, while FIG. ID illustrates further details of the cochlear implant system 102. For ease of description, FIGs. 1A-1D will generally be described together.

[0036] Cochlear implant system 102 includes an external component 104 that is configured to be directly or indirectly attached to the body of the user and an implantable component 112 configured to be implanted in the user. In the examples of FIGs. 1A-1D, the external component 104 comprises a sound processing unit 106, while the cochlear implant 112 includes an implantable coil 114, an implant body 134, and an elongate stimulating assembly 116 configured to be implanted in the user’s cochlea.

[0037] In the example of FIGs. 1A-1D, the sound processing unit 106 is an off-the-ear (OTE) sound processing unit, sometimes referred to herein as an OTE component, which is configured to send data and power to the implantable component 112. In general, an OTE sound processing unit is a component having a generally cylindrically shaped housing 111 and which is configured to be magnetically coupled to the user’s head (e.g., includes an integrated externalAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 magnet 150 configured to be magnetically coupled to an implantable magnet 152 in the implantable component 112). The OTE sound processing unit 106 also includes an integrated external (headpiece) coil 108 that is configured to be inductively coupled to the implantable coil 114.

[0038] It is to be appreciated that the OTE sound processing unit 106 is merely illustrative of the external devices that could operate with implantable component 112. For example, in alternative examples, the external component may comprise a behind-the-ear (BTE) sound processing unit or a micro-BTE sound processing unit and a separate external. In general, a BTE sound processing unit comprises a housing that is shaped to be worn on the outer ear of the user and is connected to the separate external coil assembly via a cable, where the external coil assembly is configured to be magnetically and inductively coupled to the implantable coil 114. It is also to be appreciated that alternative external components could be located in the user’s ear canal, worn on the body, etc.

[0039] As noted above, the cochlear implant system 102 includes the sound processing unit 106 and the cochlear implant 112. However, as described further below, the cochlear implant 112 can operate independently from the sound processing unit 106, for at least a period, to stimulate the user. For example, the cochlear implant 112 can operate in a first general mode, sometimes referred to as an “external hearing mode,” in which the sound processing unit 106 captures sound signals which are then used as the basis for delivering stimulation signals to the user. The cochlear implant 112 can also operate in a second general mode, sometimes referred as an “invisible hearing” mode, in which the sound processing unit 106 is unable to provide sound signals to the cochlear implant 112 (e.g., the sound processing unit 106 is not present, the sound processing unit 106 is powered-off, the sound processing unit 106 is malfunctioning, etc.). As such, in the invisible hearing mode, the cochlear implant 112 captures sound signals itself via implantable sound sensors and then uses those sound signals as the basis for delivering stimulation signals to the user. Further details regarding operation of the cochlear implant 112 in the external hearing mode are provided below, followed by details regarding operation of the cochlear implant 112 in the invisible hearing mode. It is to be appreciated that reference to the external hearing mode and the invisible hearing mode is merely illustrative and that the cochlear implant 112 could also operate in alternative modes. In FIGs. 1A and 1C, the cochlear implant system 102 is shown with an external device 110, configured to implement aspects of die techniques presented. The external device 110 is a computing device, such as a computer (e.g., laptop, desktop, tablet), a mobile phone, remote control unit, etc.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0040] In FIGs. 1A-1D, the OTE sound processing unit 106 comprises one or more input devices that are configured to receive input signals (e.g., sound or data signals). The one or more input devices include one or more sound input devices 118 (e.g., one or more external microphones, audio input ports, telecoils, etc.), one or more auxiliary input devices 128 (e.g., audio ports, such as a Direct Audio Input (DAI), data ports, such as a Universal Serial Bus (USB) port, cable port, etc.), and a wireless transmitter / receiver (transceiver) 120 (e.g., for communication with the external device 110). However, it is to be appreciated that one or more input devices may include additional types of input devices and / or less input devices (e.g., the wireless short range radio transceiver 120 and / or one or more auxiliary input devices 128 could be omitted).

[0041] The OTE sound processing unit 106 also comprises the external coil 108, a charging coil 121, a closely-coupled transmitter / receiver (RF transceiver) 122, sometimes referred to as or radio-frequency (RF) transceiver 122, at least one rechargeable battery 132, and an external sound processing module 124. The external sound processing module 124 may comprise, for example, one or more processors and a memory device (memory) that includes sound processing logic. The memory device may comprise any one or more of: Non-Volatile Memory (NVM), Ferroelectric Random Access Memory (FRAM), read only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. The one or more processors are, for example, microprocessors or microcontrollers that execute instructions for the sound processing logic stored in memory device.

[0042] The implantable component 112 comprises an implant body (main module) 134, a lead region 136, and the intra-cochlear stimulating assembly 116, all configured to be implanted under the skin / tissue (tissue) 115 of the user. The implant body 134 generally comprises a hermetically-sealed housing 138 in which RF interface circuitry 140 and a stimulator unit 142 are disposed. The implant body 134 also includes the intemal / implantable coil 114 that is generally external to the housing 138, but which is connected to the RF interface circuitry 140 via a hermetic feedthrough (not shown in FIG. ID).

[0043] As noted, stimulating assembly 116 is configured to be at least partially implanted in the user’s cochlea. Stimulating assembly 116 includes a plurality of longitudinally spaced intra-cochlear electrical stimulating contacts (electrodes) 144 that collectively form a contact or electrode array 146 for delivery of electrical stimulation (current) to the user’s cochlea.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0044] Stimulating assembly 116 extends through an opening in the user’s cochlea (e.g., cochleostomy, the round window, etc.) and has a proximal end connected to stimulator unit 142 via lead region 136 and a hermetic feedthrough (not shown in FIG. ID). Lead region 136 includes a plurality of conductors (wires) that electrically couple the electrodes 144 to the stimulator unit 142. The implantable component 112 also includes an electrode outside of the cochlea, sometimes referred to as the extra-cochlear electrode (ECE) 139.

[0045] As noted, the cochlear implant system 102 includes the external coil 108 and the implantable coil 114. The external magnet 152 is fixed relative to the external coil 108 and the implantable magnet 152 is fixed relative to the implantable coil 114. The magnets fixed relative to the external coil 108 and the implantable coil 114 facilitate the operational alignment of the external coil 108 with the implantable coil 114. This operational alignment of the coils enables the external component 104 to transmit data and power to the implantable component 112 via a closely-coupled wireless link 148 formed between the external coil 108 with the implantable coil 114. In certain examples, the closely-coupled wireless link 148 is a radio frequency (RF) link. However, various other types of energy transfer, such as infrared (IR), electromagnetic, capacitive and inductive transfer, may be used to transfer the power and / or data from an external component to an implantable component and, as such, FIG. ID illustrates only one example arrangement.

[0046] As noted above, sound processing unit 106 includes the external sound processing module 124. The external sound processing module 124 is configured to convert received input signals (received at one or more of the input devices) into output signals for use in stimulating a first ear of a user (i.e., the external sound processing module 124 is configured to perform sound processing on input signals received at the sound processing unit 106). Stated differently, the one or more processors in the external sound processing module 124 are configured to execute sound processing logic in memory to convert the received input signals into output signals that represent electrical stimulation for delivery to the user.

[0047] As noted, FIG. ID illustrates an embodiment in which the external sound processing module 124 in the sound processing unit 106 generates the output signals. In an alternative embodiment, the sound processing unit 106 can send less processed information (e.g., audio data) to the implantable component 112 and the sound processing operations (e.g., conversion of sounds to output signals) can be performed by a processor within the implantable component 112.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0048] Returning to the specific example of FIG. ID, the output signals are provided to the RF transceiver 122, which transcutaneously transfers the output signals (e.g., in an encoded manner) to the implantable component 112 via external coil 108 and implantable coil 114. That is, the output signals are received at the RF interface circuitry 140 via implantable coil 114 and provided to the stimulator unit 142. The stimulator unit 142 is configured to utilize the output signals to generate electrical stimulation signals (e.g., current signals) for delivery to the user’s cochlea. In this way, cochlear implant system 102 electrically stimulates the user’s auditory nerve cells, bypassing absent or defective hair cells that normally transduce acoustic vibrations into neural activity, in a manner that causes the user to perceive one or more components of the received sound signals.

[0049] As detailed above, in the external hearing mode the cochlear implant 112 receives processed sound signals from the sound processing unit 106. However, in the invisible hearing mode, the cochlear implant 112 is configured to capture and process sound signals for use in electrically stimulating the user’s auditory nerve cells. In particular, as shown in FIG. ID, the cochlear implant 112 includes a plurality of implantable sound sensors 160 and an implantable sound processing module 158. Similar to the external sound processing module 124, the implantable sound processing module 158 may comprise, for example, one or more processors and a memory device (memory) that includes sound processing logic. The memory device may comprise any one or more of: Non-Volatile Memory (NVM), Ferroelectric Random Access Memory (FRAM), read only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. The one or more processors are, for example, microprocessors or microcontrollers that execute instructions for the sound processing logic stored in memory device.

[0050] In the invisible hearing mode, the implantable sound sensors 160 are configured to detect / capture signals (e.g., acoustic sound signals, vibrations, etc.), which are provided to the implantable sound processing module 158. The implantable sound processing module 158 is configured to convert received input signals (received at one or more of the implantable sound sensors 160) into output signals for use in stimulating the first ear of a user (i.e., the processing module 158 is configured to perform sound processing operations). Stated differently, the one or more processors in implantable sound processing module 158 are configured to execute sound processing logic in memory to convert the received input signals into output signals 156 that are provided to the stimulator unit 142. The stimulator unit 142 is configured to utilize theAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 output signals 156 to generate electrical stimulation signals (e.g., current signals) for delivery to the user’s cochlea, thereby bypassing the absent or defective hair cells that normally transduce acoustic vibrations into neural activity.

[0051] It is to be appreciated that the above description of the so-called external hearing mode and the so-called invisible hearing mode are merely illustrative and that the cochlear implant system 102 could operate differently in different embodiments. For example, in one alternative implementation of the external hearing mode, the cochlear implant 112 could use signals captured by the sound input devices 118 and the implantable sound sensors 160 in generating stimulation signals for delivery to the user.

[0052] FIG. 2 is a functional block diagram illustrating a system 200 for generating a plurality of predicted parameters for configuring a recipient device, in accordance with certain embodiments presented herein. The recipient device can be any recipient device with configurable parameters. For example, the recipient device may be a medical device such as an implantable medical device (e.g., cochlear implant).

[0053] As illustrated in FIG. 2, the system 200 is configured to perform a plurality of operations during a training phase 262 and an inference phase 264. The training phase 262 is configured to train a machine learning model to predict a plurality of parameters for configuring a recipient device (e.g., a medical device). In operation, during the training phase 262, a parameter database 266 is accessed to obtain one or more input parameters 268. The one or more input parameters 268 may include one or more historical or current operational parameters and / or one or more historical or current recipient-specific parameters. The parameter database 266 is configured to store a plurality of operational settings or parameters for configuring one or more recipient devices. For example, a plurality of operational settings or parameters, such as stimulation settings / parameters of an implantable medical device (e.g., auditory prosthesis), may be obtained from the parameter database 266. The operational parameters may include T-level, C-level, dynamic range, frequency map, or any parameter associated with the configuration and / or stimulation of a recipient device. The operational parameters stored in the parameter database 266 may be collected from a plurality of configuration sessions for configuring one or more devices for one or more recipients (e.g., a recipient of a cochlear implant).

[0054] In addition to operational or stimulation parameters, the parameter database 266 is configured to store a plurality of recipient-specific parameters. The recipient-specificAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 parameters may include sex, age, device implantation time, device implantation time interval, electrode impedance, past operational parameters associated with a recipient, or any other parameter associated with the recipient. The recipient-specific parameters may be obtained during a configuration session of a device, or obtained through another source (e.g., medical records). In certain embodiments, the parameter database 266 may represent operational parameters associated with configuring a device for a recipient and recipient-specific parameters associated with the recipient using a vector representation. For example, each vector stored in the parameter database 266 may include operational parameters and recipientspecific parameters associated with a specific recipient. In certain embodiments, the parameter database 266 may store operational parameters and / or recipient-specific parameters associated with a specific population of recipients, such as recipients of a specific type of device (e.g., cochlear implants). The parameter database 266 may be continuously updated based as new data becomes available.

[0055] As noted above, the one or more input parameters 268 may be obtained from the parameter database 266. For example, the one or more input parameters 268 may take the form of a plurality of vectors or one or more matrices representing operational parameters and / or recipient-specific parameters. Then, the one or more input parameters 268 are provided as input to a machine learning model 270. The machine learning model 270 can include a neural network, support vector machines, K-nearest neighbors, transformer-based models, etc. Neural networks can include multi-layer perceptron or deep neural network, such as convolutional neural network, recurrent neural network, etc. The machine learning model 270 may be an unsupervised, semi-supervised, or supervised model. In certain embodiments, the machine learning model 270 may be an autoencoder network having an encoder-decoder structure, where each of the encoder and decoder is a neural network. For example, the machine learning model 270 may be a probabilistic autoencoder neural network.

[0056] Based on the one or more input parameters 268, the machine learning model 270 is configured to generate one or more predicted parameters 272. For example, the machine learning model 270 is configured to predict one or more operational parameters and / or recipient-specific parameters received in its inputs (e.g., one or more input parameters 268). That is, in other words, the machine learning model 270 is configured to reconstruct the operational parameters and / or recipient-specific parameters in the one or more input parameters 268 and provide reconstructed parameters as its outputs (e.g., the one or more predicted parameters 272). For example, the machine learning model 270 may generate the one or moreAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 predicted parameters 272 based on one or more latent variables. In certain embodiments, an input size and an output size of the machine learning model 270 may be fixed by selecting a subset of operational parameters and / or recipient-specific parameters within the one or more input parameters 268. The size of the subset of selected operational parameters and / or recipient-specific parameters may be configurable based on user requirement and / or preference.

[0057] After the one or more predicted parameters 272 (e.g., predicted T-levels and C-levels) are generated by the machine learning model 270, one or more differences between the one or more predicted parameters 272 and the one or more input parameters 268 may be determined using a loss function 274. The loss function 274 is configured to calculate a loss 276 between a predicted parameter (e.g., predicted C-level) and an input parameter (e.g., observed C-level) to determine the performance (e.g., prediction accuracy) of the machine learning model 270. For example, the loss function 274 may be an error-based loss function (e.g., mean-square error loss function or mean absolute error loss function), a binary cross entropy loss function, a categorical cross entropy loss function, or any suitable loss function. The machine learning model 270 is iteratively trained by minimizing the loss 276 generated by the loss function 274, thus resulting in continuous improvement in its predictions. In the training phase 262, the machine learning model 270 may be trained iteratively until a stopping criterion is met. The stopping criterion may include an accuracy threshold (e.g., prediction accuracy at 90% or higher), a total number of iterations, execution time, etc. The stopping criterion may be configurable based on user requirement and / or preference.

[0058] In certain embodiments, the training phase 262 may be activated to train the machine learning model 270 at certain time intervals. For example, a plurality of time intervals may be selected based on clinical data indicating when a recipient device (e.g., cochlear implant) is typically configured. By way of example, clinical data associated with cochlear implant configuration may indicate that cochlear implants are likely configured at time T1 (e.g., at the activation of the cochlear implant), at time T2 (e.g., two weeks after activation), at time T3 (e.g., one month after activation), and at time T4 (e.g., three months after activation), etc. The training phase 262 may be activated to train the machine learning model 270 at these time intervals to generate predicted parameters that track closely to actual parameters. Moreover, selecting specific time intervals to train the machine learning model 270 allows for more targeted training, thus leading to increased efficiency and accuracy.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0059] After a stopping criterion has been met, the training of the machine learning model 270 may be halted. At this point, the training phase 262 may conclude, and the inference phase 264 commences. In the inference phase 264, a latent vector 278 including a plurality of latent variables is provided as input to a trained machine learning model 280. In certain embodiments, one or more matrices representing a plurality of latent vectors (e.g., latent vector 278) may be provided as input to the trained machine learning model 280. In certain embodiments, the trained machine learning model 280 may take the form of the machine learning model 270 after it has been trained to meet a stopping criterion as noted above.

[0060] In certain embodiments, the latent vector 278 may include a plurality of latent variables that are randomly generated or initialized. For example, the plurality of latent variables may be randomly sampled from a latent space using a Monte Carlo method or any suitable sampling method. In certain embodiments, the plurality of latent variables may be initialized following a grid. Taking the latent vector 278 as input, the trained machine learning model 280 generates one or more predicted parameters 282 (e.g., predicted T-levels and C-levels). Then, one or more differences between the one or more predicted parameters 282 and the one or more observed parameters 284 (which may include at least one new observation 286) are determined using a loss function 288. The loss function 288 is configured to calculate a loss 290 between a predicted parameter (e.g., predicted C-level) and an observed parameter (e.g., observed C- level) based on a new observation (e.g., measurement made by a medical practitioner) to determine the performance (e.g., prediction accuracy) of the trained machine learning model 280 and its corresponding latent vector 278. For example, the loss function 288 may be an error-based loss function (e.g., mean-square error loss function or mean absolute error loss function), binary cross entropy loss function, categorical cross entropy loss function, or any suitable loss function.

[0061] In certain embodiments, a cost function may determine a cost as an average of the outputs (e.g., loss 290) of the loss functions over an entire input data. The cost may be minimized iteratively by backpropagating it through the trained machine learning model 280 while optimizing the plurality of latent variables in latent vector 278. That is, in other words, the plurality of latent variables is optimized to identify a set of latent variables that best match the observed and predicted parameters. The optimization process may continue until a stopping criterion is met. The stopping criterion may include an accuracy threshold (e.g., prediction accuracy is 90% or higher), a total number of iterations, execution time, etc. The stopping criterion may be configurable based on user requirement and / or preference.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0062] In certain embodiments, the trained machine learning model 280 may generate predictions of both observed and unobserved parameters based on the optimized latent variables. As the trained machine learning model 280 remains fixed during the inference phase 264 (e.g., parameters, such as layer weights, of the trained machine learning model 280 are unchanged), the predicted unobserved parameter values are meaningful and correspond to data of a recipient population (e.g., cochlear implant recipients) provided by a database (e.g., parameter database 266) or another source . In certain embodiments, operations of the inference phase 264 may be repeated each time anew observation (e.g., new observation 286) is provided by a source (e.g., a medical practitioner).

[0063] FIG. 3 is a functional block diagram illustrating a system 300 for generating a plurality of predicted parameters for configuring a recipient device, in accordance with certain embodiments presented herein. As noted above, the recipient device can be any recipient device with configurable parameters. For example, the recipient device may be a medical device such as an implantable medical device (e.g., cochlear implant).

[0064] As illustrated in FIG. 3, the system 300 is configured to perform a plurality of operations during a training phase 362 and an inference phase 364. The training phase 362 is configured to train an autoencoder network to predict a plurality of parameters for configuring a recipient device (e.g., a medical device). In operation, during the training phase 362, a parameter database 366 is accessed to obtain one or more input parameters 368. The one or more input parameters 368 may one or more historical or current operational parameters and / or one or more historical or current recipient-specific parameters. The parameter database 366 is configured to store a plurality of operational settings or parameters for configuring one or more recipient devices. For example, a plurality of operational settings or parameters, such as stimulation settings / parameters of an implantable medical device (e.g., auditory prosthesis), may be obtained from the parameter database 366. The operational parameters may include T- level, C-level, dynamic range, frequency map, or any parameter associated with the configuration and / or stimulation of a recipient device. The operational parameters stored in the parameter database 366 may be collected from a plurality of configuration sessions for configuring one or more devices for one or more recipients (e.g., a recipient of a cochlear implant).

[0065] In addition to operational or stimulation parameters, the parameter database 366 is configured to store a plurality of recipient-specific parameters. The recipient-specific parameters may include sex, age, device implantation time, device implantation time interval,Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 electrode impedance, past operational parameters associated with a recipient, or any other parameter associated with the recipient. The recipient-specific parameters may be obtained during a configuration session of a recipient device, or obtained through another source (e.g., medical records). In certain embodiments, the parameter database 366 may represent operational parameters associated with configuring a recipient device for a recipient and the recipient-specific parameters associated with the recipient using a vector representation. For example, each vector stored in the parameter database 366 may include operational parameters and recipient-specific parameters associated with a specific recipient. In certain embodiments, the parameter database 366 may store operational parameters and / or recipient-specific parameters associated with a specific population of recipients, such as recipients of a specific type of device (e.g., cochlear implants). The parameter database 366 may be continuously updated based as new data becomes available.

[0066] As noted above, the one or more input parameters 368 may be obtained from the parameter database 366. For example, the one or more input parameters 368 may take the form of a plurality of vectors or one or more matrices representing operational parameters and / or recipient-specific parameters. Then, the one or more input parameters 368 are provided as input to an autoencoder network 370. The autoencoder network 370 includes an encoder network 371 configured to map the one or more input parameters 368 into a feature vector 372 (e.g., a feature vector in a latent space). In certain embodiments, one or more matrices representing a plurality of feature vectors (e.g., feature vector 372) may be provided as input to the decoder network 373. The autoencoder network 370 further includes a decoder network 373 configured to reconstruct the one or more input parameters 368 based on the feature vector 372. In certain embodiments, the autoencoder network 370 may be implemented as an autoencoder neural network where each of the encoder network 371 and decoder network 373 may be a neural network, such as a feed forward network, convolutional neural network, recurrent neural network, long short-term memory (LSTM) network, etc. In certain embodiments, the autoencoder network 370 may be implemented as a generative adversarial network (GAN) with the encoder network 371 as a generator and the decoder network 373 as a discriminator.

[0067] For example, the decoder network 373 may be a neural network having a plurality of layers, where the deepest layer includes a plurality of latent variables configured to reconstruct the operational parameters and / or recipient-specific parameters of the one or more input parameters 368 based on the feature vector 372. Based on the reconstruction, the autoencoderAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 network 370 is configured to generate one or more predicted parameters 374. That is, in other words, the autoencoder network 370 is configured to reconstruct the one or more input parameters 368 and output the reconstructed parameters as the one or more predicted parameters 374. In certain embodiments, an input size and an output size of the autoencoder network 370 may be fixed by selecting a subset of operational parameters and / or recipientspecific parameters within the one or more input parameters 368. The size of the subset of selected operational parameters and / or recipient-specific parameters may be configurable based on user requirement and / or preference.

[0068] After the one or more predicted parameters 374 (e.g., predicted T-levels and C-levels) are generated by the autoencoder network 370, one or more differences between the one or more predicted parameters 374 and the one or more input parameters 368 may be determined using a loss function. The loss function may be an error-based loss function 375 configured to calculate a loss as an error in the predictions generated by the autoencoder network 370. That is, for example, the error-based loss function 375 is configured to calculate the difference between a predicted parameter (e.g., predicted C-level) and an input parameter (e.g., observed C-level) to determine the performance (e.g., prediction accuracy) of the autoencoder network 370. For example, the error-based loss function 375 may be a mean-square error loss function or mean absolute error loss function. In certain embodiments, the loss function may be entropybased, such as a binary cross entropy loss function, categorical cross entropy loss function, or any suitable loss function. The autoencoder network 370 is iteratively trained by minimizing the loss generated by the error-based loss function 375, thus resulting in continuous improvement of predictions generated by the autoencoder network 370. In the training phase 362, the autoencoder network 370 may be trained iteratively until a stopping criterion is met. The stopping criterion may include an accuracy threshold (e.g., prediction accuracy must be 90% or higher), a total number of iterations, execution time, etc. The stopping criterion may be configurable based on user requirement and / or preference.

[0069] In certain embodiments, the training phase 362 may be activated to train the autoencoder network 370 at certain time intervals. For example, a plurality of time intervals may be selected based on clinical data indicating when a recipient device (e.g., cochlear implant) is typically configured. By way of example, clinical data associated with cochlear implant configuration may indicate that cochlear implants are likely configured at time T 1 (e .g . , at the activation of the cochlear implant), at time T2 (e.g., two weeks after activation), at time T3 (e.g., one month after activation), and at time T4 (e.g., three months after activation), etc.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1The training phase 362 may be activated to train the autoencoder network 370 at these time intervals to generate predicted parameters that track closely to actual parameters. Moreover, selecting specific time intervals for training the autoencoder network 370 allows for more targeted training, thus leading to increased efficiency and accuracy.

[0070] After a stopping criterion has been met, the training of the autoencoder network 370 may be halted. At this point, the system 300 is configured to freeze the decoder network 373 by fixing its parameters (e.g., layer weights) at an operation 376. The decoder network 373, which has fixed-length input and output layers, may serve as the main structure by which variable -length unobserved parameters (e.g., unobserved operational parameters) are estimated / predicted. After the operation 376 concludes, the inference phase 364 may commence. In the inference phase 364, a latent vector 378 including a plurality of latent variables may be provided as input to a trained decoder network 380. In certain embodiments, one or more matrices representing a plurality of latent vectors (e.g., latent vector 378) may be provided as input to the trained decoder network 380. In certain embodiments, the trained decoder network 380 may be the decoder network 373 after it has been trained to meet a stopping criterion as noted above and after it has been “frozen” with fixed parameters (e.g., fixed layer weights) at the operation 376.

[0071] In certain embodiments, the latent vector 378 may include a plurality of latent variables that are randomly generated or initialized. For example, the plurality of latent variables may be randomly sampled from a latent space using a Monte Carlo method or any suitable sampling method. In certain embodiments, the plurality of latent variables may be initialized following a grid. Taking the latent vector 378 as input, the trained decoder network 380 generates one or more predicted parameters 382 (e.g., predicted T-levels and C-levels). The one or more predicted parameters 382 and one or more observed parameters 384 (which may include at least one new observation 386) may be compared to determine a prediction accuracy of the trained decoder network 380. For example, at an operation 388, one or more differences may be determined between the one or more predicted parameters 382 and the one or more observed parameters 384. The one or more observed parameters 384 include one or more observed operational parameters, such as the new observation 386 (e.g., observed C-level or T-level) made by a medical practitioner during a device configuration session.

[0072] The one or more differences between the one or more predicted parameters 382 and the one or more observed parameters 384 may be determined using a loss function. The loss function is configured to calculate a loss between a predicted parameter (e.g., predicted C-Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 level) and an observed parameter (e.g., observed C-level) based on a new observation (e.g., measurement made by a medical practitioner) to determine the performance (e.g., prediction accuracy) of the trained decoder network 380 and its corresponding latent vector 378. For example, the loss function may be an error-based loss function (e.g., mean-square error loss function or mean absolute error loss function), binary cross entropy loss function, categorical cross entropy loss function, or any suitable loss function.

[0073] In certain embodiments, a cost function may determine a cost as an average of the outputs (e.g., loss) of the loss functions over an entire input data. At operation 390, the latent vector 378 may be updated through gradient descent based on the one or more differences computed at operation 388. The cost / loss may be minimized iteratively by backpropagating it through the trained decoder network 380 while optimizing the plurality of latent variables in latent vector 378. That is, in other words, the plurality of latent variables is optimized to identify a set of latent variables that best match the observed and predicted parameters. The optimization process may continue until a stopping criterion is met. The stopping criterion may include an accuracy threshold (e.g., prediction accuracy must be 90% or higher), a total number of iterations, execution time, etc. The stopping criterion may be configurable based on user requirement and / or preference.

[0074] In certain embodiments, the trained decoder network 380 may generate predictions of both observed and unobserved parameters based on the optimized latent variables. As the trained decoder network 380 remains fixed during the inference phase 364 (e.g., parameters, such as weights, of the trained decoder network 380 are unchanged), the predicted unobserved parameter values are meaningful and correspond to data of a recipient population (e.g., cochlear implant recipients) provided by a database (e.g., parameter database 366) or another source. In certain embodiments, operations of the inference phase 364 may be repeated each time a new observation (e.g., new observation 386) is provided by a source (e.g., a medical practitioner).

[0075] FIG. 4 is a functional block diagram illustrating a system 400 for generating, via a trained decoder network, a plurality of predicted parameters for configuring a recipient device, in accordance with certain embodiments presented herein. As noted above, the recipient device can be any recipient device with configurable parameters. For example, the recipient device may be a medical device such as an implantable medical device (e.g., cochlear implant).

[0076] As illustrated in FIG. 4, the system 400 is configured to perform a plurality of operations during an inference phase 464. In the inference phase 464, a latent vector 478 mayAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 be provided as input to a trained decoder network 480. In certain embodiments, one or more matrices representing a plurality of latent vectors (e.g., latent vector 478) may be provided as input to the trained decoder network 480. As noted above, the trained decoder network 480 may be a decoder network that has been trained to meet a stopping criterion and “frozen” with fixed parameters (e.g., fixed layer weights). In certain embodiments, the latent vector 478 may include a plurality of latent variables that are randomly generated or initialized. For example, the plurality of latent variables may be randomly sampled from a latent space using a Monte Carlo method or any suitable sampling method. In certain embodiments, the plurality of latent variables may be initialized following a grid.

[0077] Taking the latent vector 478 as input, the trained decoder network 480 generates one or more predicted parameters 482 (e.g., predicted T-levels and C-levels). The one or more predicted parameters 482 and one or more observed parameters 484 may be compared to determine a prediction accuracy of the trained decoder network 380. The one or more observed parameters 484 may include one or more historical observed parameters obtained from a plurality of past configuration sessions, such as a past session at time T1 485A, a past session at time T2 485B, . . . , or a past session at Tn 485N. Alternatively or additionally, the one or more observed parameters 484 may include one or more observed parameters based on a new observation 486 made by a medical practitioner or obtained from another source. Alternatively or additionally, the one or more observed parameters 484 may also include one or more observed parameters obtained from a current configuration session conducted by a medical practitioner. In certain embodiments, certain observed parameters may not be available, such as observed parameters associated with the past session at time T2 485B.

[0078] After obtaining the one or more observed parameters 484, one or more differences may be determined between the one or more predicted parameters 482 and the one or more observed parameters 484 at operation 488. For example, the differences may be determined using a loss function configured to calculate a loss between a predicted parameter (e.g., predicted C-level) and an observed parameter (e.g., observed C-level) to determine the performance (e.g., prediction accuracy) of the trained decoder network 480 and its corresponding latent vector 478. For example, the loss function may be an error-based loss function (e.g., mean-square error loss function or mean absolute error loss function), binary cross entropy loss function, categorical cross entropy loss function, or any suitable loss function. In certain embodiments, a cost function may determine a cost as an average of the outputs (e.g., loss) of the loss functions over an entire input data. In certain embodiments, the cost / loss may be determinedAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 only based on available observed data. For example, observed parameters associated with the past session at time T2485B are not available, and thus data associated with this session would not be used in the cost / loss determination.

[0079] At operation 490, the latent vector 478 may be updated through gradient descent based on the one or more differences computed at operation 488. The cost / loss may be minimized iteratively by backpropagating it through the trained decoder network 480 while optimizing the plurality of latent variables in latent vector 478. That is, in other words, the plurality of latent variables is optimized to identify a set of latent variables that best match the observed and predicted parameters. The optimization process may continue until a stopping criterion has been met. The stopping criterion may include an accuracy threshold (e.g., prediction accuracy must be 90% or higher), a total number of iterations, execution time, etc. The stopping criterion may be configurable based on user requirement and / or preference.

[0080] In certain embodiments, the trained decoder network 480 may generate predictions of both observed and unobserved parameters based on the optimized latent variables. As the trained decoder network 480 remains fixed during the inference phase 464 (e.g., parameters, such as weights, of the trained decoder network 480 are unchanged), the predicted unobserved parameter values are meaningful and correspond to data of a recipient population (e.g., cochlear implant recipients) provided by a database or another source. In certain embodiments, operations of the inference phase 464 may be repeated each time a new observation (e.g., new observation 486) has been provided by a source (e.g., a medical practitioner).

[0081] FIG. 5 is a functional block diagram illustrating a system 500 for determining uncertainty values associated with a plurality of predicted parameters, in accordance with certain embodiments presented herein. Initially, at operation 577, a plurality of latent variables may be randomly sampled from a latent space based on trained characteristics of the latent variables. In certain embodiments, the trained characteristics may be expressed by a Gaussian multivariate distribution that is identified by a mean vector and a covariance matrix. For example, the plurality of latent variables, represented as a plurality of latent vectors 578, may be randomly sampled from a latent space using any suitable sampling method, including a Monte Carlo method. The number of iterations to run a sampling method (e.g., Monte Carlo method) may depend on a desired accuracy and / or other predetermined requirements (e.g., time or resource constraints). After running the sampling method for a number of iterations (e.g., 10,000 iterations), the plurality of latent vectors 578 are generated and provided as input to a trained decoder network 580. As noted above, the trained decoder network 580 may be aAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 decoder network that has been trained to meet a stopping criterion and “frozen” with fixed parameters (e.g., fixed layer weights).

[0082] Based on the plurality of latent vectors 578, the trained decoder network 580 is configured to generate output vectors containing predicted parameters 582. Then, at operation 588, one or more differences between each of the output vectors containing predicted parameters 582 and a vector containing a plurality of observed parameters 586 are determined. For example, the one or more differences may be determined using a loss function. The loss function is configured to calculate a loss between a predicted parameter (e.g., predicted C- level) and an observed parameter (e.g., observed C-level) to calculate a probability distribution related to parameter uncertainty. For example, the loss function may be an error-based loss function (e.g., mean-square error loss function or mean absolute error loss function), binary cross entropy loss function, categorical cross entropy loss function, or any suitable loss function.

[0083] Based on the one or more differences (e.g., mean-square error values), output vectors containing predicted parameters 582 are evaluated. Each output vector of predicted parameters that provides a good match to the plurality of observed parameters 586 is selected and added to a dictionary or any suitable data structure. Otherwise, the output vector is considered rejected. That is, at operation 589, output vectors containing predicted parameters having differences with the plurality of observed parameters 586 (e.g., mean-square error values) below a threshold of acceptance (e.g., an error threshold) may be selected and added into the dictionary. Then, a probability distribution may be calculated for each output vector in the dictionary, wherein the probability distribution describes the likelihood of possible values for the predicted parameters in the output vector. At operation 591, a mean and a variance associated with each probability distribution is computed for each predicted parameter of the selected output vectors. In certain embodiments, in addition to mean and variance, other statistical parameters such as standard deviation may be calculated.

[0084] The computed mean and variance values may be used to determine an uncertainty of the predicted parameters. For example, a high variance value may indicate high uncertainty associated with a particular prediction by the trained decoder network 580. Each of the resulting probability distributions expresses a confidence level that a user (e.g., a medical practitioner) should have in the predicted unobserved and / or observed parameters (e.g., operational parameters). For example, a wide probability distribution indicates greater variance (e.g., greater uncertainty). Thus, the wider the probability distribution, the lessAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 confident the user should be about a corresponding predicted parameter. Based on the uncertainty values (e.g., variance values) associated with the predicted parameters, the user may select a subset of predicted parameters having the greatest variance for a next round of observation. Overall, the uncertainty values allow users to quickly determine which parameters predicted by the trained decoder network 580 are reliable and which parameters require further clinical observation, thus providing increased efficiency and accuracy in device configuration sessions.

[0085] FIG. 6 is a graphical depiction illustrating a graph 600 for visualizing uncertainty values associated with a plurality of predicted parameters, in accordance with certain embodiments presented herein. The graph 600 includes an x-axis representing an electrode number 692 identifying an electrode of a recipient device (e.g., electrode number 1 through 22 of a cochlear implant). The graph 600 further includes a y-axis representing an operational parameter 693 (e.g., C-level or T-level) corresponding to each of the electrodes. Each electrode number 692 corresponds with each of a plurality of predicted points 694 indicating a predicted value for the operational parameter 693. For example, the plurality of predicted points 694 may be predicted operational parameters (e.g., predicted C-levels) generated by a machine learning model (e.g., a trained decoder network of an autoencoder network). The plurality of predicted points 694 may be connected to form a line representing predicted operational parameters for a plurality of electrodes.

[0086] All or some of the electrodes may correspond with at least one of a plurality of measured points 695. Each of the plurality of measured points 695 may include an actual measurement (e.g., observation) made by a user (e.g., a medical practitioner) or obtained from another source. For example, electrode number 1 may be associated with a predicted point (of the plurality of predicted points 694) and a measured point (of the plurality of measured points 695), while electrode number 2 may only be associated with a predicted point (of the plurality of predicted points 694). In other words, an observation may not be available for electrode number 2. Based on the plurality of measured points 695 (e.g., observed data points), a default linear interpolation 696 may be established to estimate unobserved data points (e.g., unobserved operational parameter for electrode number 2).

[0087] Further, a probability distribution is determined for each of the plurality of predicted points 694. Then, a mean and a variance of the probability distribution may be computed for each of the plurality of predicted points 694. In certain embodiments, in addition to mean and variance, other statistical parameters such as standard deviation may be calculated. TheAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 computed mean and variance values may be used to determine prediction uncertainty 697 for the predicted parameters. For example, a high variance value may indicate high prediction uncertainty associated with a particular prediction.

[0088] The prediction uncertainty 697 may be visualized as a vertical bar representing a width / spread of the probability distribution. For example, each of the resulting probability distributions expresses a confidence level that a user (e.g., a medical practitioner) should have in the predicted unobserved and / or observed parameters (e.g., operational parameters). For example, a wide probability distribution indicates greater variance (e.g., greater uncertainty) in the plurality of predicted points 694. Thus, the wider the probability distribution, the less confident the user should be about a corresponding predicted parameter. Based on the uncertainty values (e.g., variance values) associated with the predicted parameters, the user may select a subset of predicted parameters having the greatest variance for a next round of observation. For example, one or more electrodes having parameters with variance values above a predetermined threshold may be selected for further observation. This way, the uncertainty values allow users (e.g., a medical practitioner) to quickly determine which of the plurality of predicted points 694 is reliable and which parameters require further clinical observation, thus providing increased efficiency and accuracy in the configuration of recipient devices.

[0089] In certain embodiments, the graph 600 may be displayed, via a graphical user interface (GUI), to a user (e.g., a medical practitioner) before, during, or after a recipient device configuration session. Based on the prediction uncertainty 697 (e.g., variance values) visualized in graph 600, the user may quickly and accurately decide which electrode needs further clinical observation (measurement) at a current or future session. For example, the user may speed up the recipient device configuration session by only obtaining new observations (measurements) of operational parameters for electrodes with prediction uncertainty 697 higher than a predetermined threshold while relying on predicted parameters for the other electrodes. This provides a more targeted configuration experience, thus saving time and resources for the recipient and the medical practitioner. Further, augmenting model-generated parameter predictions with human observation and expertise ensures the recipient device is properly configured. In certain embodiments, the graphical user interface may be configured to receive input from the user (e.g., medical practitioner), such as feedback regarding the plurality of predicted points 694 and / or prediction uncertainty 697.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0090] By way of example, the user may review the graph 600 to prepare for an upcoming configuration session. Based on expert knowledge, review of medical records, and / or information previously obtained from the recipient, the user may determine that a predicted point for an operational parameter (e.g., C-level) of a specific electrode is inaccurate. The user may provide feedback on the parameter predictions via the GUI using any suitable method of input. For example, the user may interact with the graph 600 via the GUI using a drag-and- drop technique. That is, for example, the user may drag an inaccurate predicted point and drop it to a desired location on the graph 600 (e.g., moving a predicted point associated with electrode 3 from C-level = 30 to C-level = 40). In certain embodiments, feedback from the user may be used to refine and / or update the machine learning model (e.g., autoencoder network) that generated the plurality of predicted points 694. The user feedback can help finetune the machine learning model to generate more accurate predictions, thus leading to a more robust device configuration experience for both the user and the device recipient.

[0091] FIG. 7 is a flowchart illustrating a method 700, in accordance with certain embodiments presented herein. Method 700 begins at 701 where a plurality of predicted operational parameters of an implantable medical device is generated via a decoder of a trained autoencoder network based on a latent vector having a plurality of latent variables. At 703, a difference between the plurality of predicted operational parameters and a plurality of observed operational parameters of the implantable medical device is iteratively backpropagated to generate an updated latent vector. Then, at 705, based on the updated latent vector, a plurality of operational parameters is generated via the decoder. At 707, the plurality of operational parameters is configured with the plurality of operational parameters. As shown by dashed arrow 709, the operations of FIG. 7 can be repeated at various times (e.g., within a single device fitting session, during different device fitting sessions, etc.).

[0092] FIG. 8 is a flowchart illustrating a method 800, in accordance with certain embodiments presented herein. Method 800 begins at 801 where a latent space is sampled to obtain one or more latent vectors each having a plurality of latent variables. At 803, one or more vectors, each having a plurality of predicted stimulation parameters for a recipient device based the one or more latent vectors, are generated via a decoder of a trained autoencoder network. Then, at 805, a prediction uncertainty of the one or more vectors is determined. At 807, one or more stimulation parameters for the recipient device are configured based on the prediction uncertainty.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0093] FIG. 9 is a flowchart illustrating a method 900, in accordance with certain embodiments presented herein, where the operations are performed by a system comprising a memory and at least one processor operable coupled to the memory. At 901, the processor is configured to generate, via a decoder of a trained autoencoder network, a plurality of predicted unobserved stimulation parameters for a recipient device. At 903, the processor is configured to select one or more electrodes of the recipient device to be configured based on the plurality of predicted unobserved stimulation parameters. Then, at 905, the processor is configured to configure one or more unobserved stimulation parameters of the one or more electrodes.

[0094] FIG. 10 is a flowchart illustrating a method 1000, in accordance with certain embodiments presented herein, where the operations are performed by a processor executing instructions stored in one or more non-transitory computer readable storage media. At 1001, a plurality of operational parameters is generated via atrained machine learning model. At 1003, a prediction uncertainty of the plurality of operational parameters is determined. Then, at 1005, one or more electrodes of a recipient device based on the prediction uncertainty are selected based on the prediction uncertainty. At 1007, the one or more electrodes are configured based on the plurality of operational parameters.

[0095] FIG. 11 is a flowchart illustrating a method 1100, in accordance with certain embodiments presented herein, where the operations are performed by a system comprising a memory and at least one processor operable coupled to the memory. At 1101, the processor is configured to obtain one or more latent vectors each having a plurality of latent variables. At 1103, the processor is configured to generate, via a decoder of a trained autoencoder network, one or more predicted vectors each having a plurality of predicted stimulation parameters for a device based the one or more latent vectors. Then, at 1105, the processor is configured to configure one or more stimulation parameters for the device based on the plurality of predicted stimulation parameters.

[0096] As previously described, the technology disclosed herein can be applied in any of a variety of circumstances and with a variety of different devices. Example devices that can benefit from technology disclosed herein are described in more detail in FIGS. 12 and 13. The techniques of the present disclosure can be applied to other devices, such as neurostimulators, cardiac pacemakers, cardiac defibrillators, sleep apnea management stimulators, seizure therapy stimulators, tinnitus management stimulators, and vestibular stimulation devices, as well as other medical devices that deliver stimulation to tissue. Further, technology describedAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 herein can also be applied to consumer devices. These different systems and devices can benefit from the technology described herein.

[0097] FIG. 12 illustrates an example vestibular stimulator system 1202, with which embodiments presented herein can be implemented. As shown, the vestibular stimulator system 1202 comprises an implantable component (vestibular stimulator) 1212 and an external device / component 1204 (e.g., external processing device, battery charger, remote control, etc.). The external device 1204 comprises a transceiver unit 1260. As such, the external device 1204 is configured to transfer data (and potentially power) to the vestibular stimulator 1212.

[0098] The vestibular stimulator 1212 comprises an implant body (main module) 1234, a lead region 1236, and a stimulating assembly 1216, all configured to be implanted under the skin / tissue (tissue) 1215 of the recipient. The implant body 1234 generally comprises a hermetically-sealed housing 1238 in which RF interface circuitry, one or more rechargeable batteries, one or more processors, and a stimulator unit are disposed. The implant body 1234 also includes an intemal / implantable coil 1214 that is generally external to the housing 1238, but which is connected to the transceiver via a hermetic feedthrough (not shown).

[0099] The stimulating assembly 1216 comprises a plurality of electrodes 1244(l)-(3) disposed in a carrier member (e.g., a flexible silicone body). In this specific example, the stimulating assembly 1216 comprises three (3) stimulation electrodes, referred to as stimulation electrodes 1244(1), 1244(2), and 1244(3). The stimulation electrodes 1244(1), 1244(2), and 1244(3) function as an electrical interface for delivery of electrical stimulation signals to the recipient’s vestibular system.[ootoo] The stimulating assembly 1216 is configured such that a surgeon can implant the stimulating assembly adjacent the recipient’s otolith organs via, for example, the recipient’s oval window. It is to be appreciated that this specific embodiment with three stimulation electrodes is merely illustrative and that the techniques presented herein may be used with stimulating assemblies having different numbers of stimulation electrodes, stimulating assemblies having different lengths, etc.[ooiot] In operation, the vestibular stimulator 1212, the external device 1204, and / or another external device can be configured to implement the techniques presented herein. That is, the vestibular stimulator 1212, possibly in combination with the external device 1204 and / or another external device, can include an evoked biological response analysis system, as described elsewhere herein.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0102] FIG. 13 illustrates a retinal prosthesis system 1301 that comprises an external device 1310 (which can correspond to the wearable device 100) configured to communicate with an implantable retinal prosthesis 1300 via signals 1351. The retinal prosthesis 1300 comprises an implanted processing module 1325, and a retinal prosthesis sensor-stimulator 1390 is positioned proximate the retina of a recipient. The external device 1310 and the processing module 1325 can communicate via coils 1308, 1314.

[0103] In an example, sensory inputs (e.g., photons entering the eye) are absorbed by a microelectronic array of the sensor-stimulator 1390 that is hybridized to a glass piece 1392 including, for example, an embedded array of microwires. The glass can have a curved surface that conforms to the inner radius of the retina. The sensor-stimulator 1390 can include a microelectronic imaging device that can be made of thin silicon containing integrated circuitry that convert the incident photons to an electronic charge.

[0104] The processing module 1325 includes an image processor 1323 that is in signal communication with the sensor-stimulator 1390 via, for example, a lead 1388 that extends through surgical incision 1389 formed in the eye wall. In other examples, processing module 1325 is in wireless communication with the sensor-stimulator 1390. The image processor 1323 processes the input into the sensor-stimulator 1390 and provides control signals back to the sensor-stimulator 1390 so the device can provide an output to the optic nerve. That said, in an alternate example, the processing is executed by a component proximate to, or integrated with, the sensor-stimulator 1390. The electric charge resulting from the conversion of the incident photons is converted to a proportional amount of electronic current which is input to a nearby retinal cell layer. The cells fire and a signal is sent to the optic nerve, thus inducing a sight perception.

[0105] The processing module 1325 can be implanted in the recipient and function by communicating with the external device 1310, such as a BTE unit, a pair of eyeglasses, etc. The external device 1310 can include an external light / image capture device (e.g., located in / on a behind-the-ear device or a pair of glasses, etc.), while, as noted above, in some examples, the sensor-stimulator 1390 captures light / images, in which sensor-stimulator 1390 is implanted in the recipient.

[0106] Turning to FIG. 14, depicted therein is a block diagram illustrating an example fitting system 1470 configured to perform aspects of the techniques presented herein (e.g., perform one or more of the operations of FIGs. 2, 3, 4, 5, 6, 7, 8, 9, 10, and / or 11). Fitting system 1470Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 is, in general, a computing device that comprises a plurality of interfaces / ports 1478(1)- 1478(N), a memory 1480, a processor 1484, and a user interface 1486. The interfaces 1478(1)- 1478(N) may comprise, for example, any combination of network ports (e.g., Ethernet ports), wireless network interfaces, Universal Serial Bus (USB) ports, Institute of Electrical and Electronics Engineers (IEEE) 1394 interfaces, PS / 2 ports, etc. In the example of FIG. 14, interface 1478(1) is connected to cochlear implant system 102 having components implanted in a user 1471. Interface 1478(1) may be directly connected to the cochlear implant system 102 or connected to an external device that is communication with the cochlear implant systems (e.g., external device 110 of FIGs. 1A-1D). Interface 1478(1) may be configured to communicate with cochlear implant system 102 via a wired or wireless connection (e.g., telemetry, Bluetooth, etc.).

[0107] The user interface 1486 includes one or more output devices, such as a display screen (e.g., a liquid crystal display (LCD)) and a speaker, for presentation of visual or audible information to a clinician, audiologist, or other user. The user interface 1486 may also comprise one or more input devices that include, for example, a keypad, keyboard, mouse, touchscreen, etc.

[0108] The memory 1480 comprises latent variable fitting logic 1481 that may be executed to perform aspects of the techniques presented herein (e.g., executed to perform one or more of the operations of FIGs. 2, 3, 4, 5, 6, 7, 8, 9, 10, and / or 11). Memory 1480 may comprise read only memory (ROM), random access memory (RAM), magnetic disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. The processor 1484 is, for example, a microprocessor or microcontroller that executes instructions for the latent variable fitting logic 1481. Thus, in general, the memory 1480 may comprise one or more tangible (non-transitory) computer readable storage media (e.g., a memory device) encoded with software comprising computer executable instructions and when the software is executed (by the processor 1484) it is operable to perform the techniques described herein.

[0109] As should be appreciated, while particular uses of the technology have been illustrated and discussed above, the disclosed technology can be used with a variety of devices in accordance with many examples of the technology. The above discussion is not meant to suggest that the disclosed technology is only suitable for implementation within systems akin to that illustrated in the figures. In general, additional configurations can be used to practiceAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 the processes and systems herein and / or some aspects described can be excluded without departing from the processes and systems disclosed herein.[ooito] This disclosure described some aspects of the present technology with reference to the accompanying drawings, in which only some of the possible aspects were shown. Other aspects can, however, be embodied in many different forms and should not be construed as limited to the aspects set forth herein. Rather, these aspects were provided so that this disclosure was thorough and complete and fully conveyed the scope of the possible aspects to those skilled in the art.[oom] As should be appreciated, the various aspects (e.g., portions, components, etc.) described with respect to the figures herein are not intended to limit the systems and processes to the particular aspects described. Accordingly, additional configurations can be used to practice the methods and systems herein and / or some aspects described can be excluded without departing from the methods and systems disclosed herein.

[0112] According to certain aspects, systems and non-transitory computer readable storage media are provided. The systems are configured with hardware configured to execute operations analogous to the methods of the present disclosure. The one or more non-transitory computer readable storage media comprise instructions that, when executed by one or more processors, cause the one or more processors to execute operations analogous to the methods of the present disclosure.

[0113] Similarly, where steps of a process are disclosed, those steps are described for purposes of illustrating the present methods and systems and are not intended to limit the disclosure to a particular sequence of steps. For example, the steps can be performed in differing order, two or more steps can be performed concurrently, additional steps can be performed, and disclosed steps can be excluded without departing from the present disclosure. Further, the disclosed processes can be repeated.

[0114] Although specific aspects were described herein, the scope of the technology is not limited to those specific aspects. One skilled in the art will recognize other aspects or improvements that are within the scope of the present technology. Therefore, the specific structure, acts, or media are disclosed only as illustrative aspects. The scope of the technology is defined by the following claims and any equivalents therein.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1

[0115] It is also to be appreciated that the embodiments presented herein are not mutually exclusive and that the various embodiments may be combined with another in any of a number of different manners.

Claims

1. Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1CLAIMSWhat is claimed is:

1. A method comprising: generating, via a decoder of a trained autoencoder network, a plurality of predicted operational parameters of an implantable medical device based on a latent vector having a plurality of latent variables; iteratively backpropagating a difference between the plurality of predicted operational parameters and a plurality of observed operational parameters of the implantable medical device to generate an updated latent vector; generating, via the decoder, a plurality of operational parameters based on the updated latent vector; and configuring the implantable medical device with the plurality of operational parameters.

2. The method of claim 1, wherein one or more weights of the decoder are fixed.

3. The method of claim 1, wherein the plurality of latent variables is randomly initialized.

4. The method of claim 1, wherein iteratively backpropagating the difference between the plurality of predicted operational parameters and the plurality of observed operational parameters of the implantable medical device to generate the updated latent vector comprises: iteratively updating the plurality of latent variables to minimize a cost function measuring the difference, wherein the cost function includes a mean-square error between the plurality of predicted operational parameters and the plurality of observed operational parameters.

5. The method of claim 1, wherein the plurality of operational parameters includes observed and unobserved parameters for configuring the implantable medical device.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC16. The method of claim 1, 2, 3, 4, or 5, wherein the trained autoencoder network is generated based on training an autoencoder network with a plurality of historical operational parameters and a plurality of recipient-specific parameters.

7. The method of claim 6. wherein the plurality of historical operational parameters is associated with one or more implantable medical devices each having a plurality of electrodes, and wherein the plurality of historical operational parameters comprises one or more of: threshold levels associated with the plurality of electrodes, comfort levels associated with the plurality of electrodes, or frequency maps associated with the plurality of electrodes.

8. The method of claim 6, wherein the plurality of recipient-specific parameters comprises one or more of: sex, age, implantation time associated with the implantable medical device, electrode impedance values associated with a plurality of electrodes of the implantable medical device, or a plurality of recipient-specific operational parameters.

9. A method comprising: sampling a latent space to obtain one or more latent vectors each having a plurality of latent variables; generating, via a decoder of a trained autoencoder network, one or more vectors each having a plurality of predicted stimulation parameters for a recipient device based on the one or more latent vectors; determining a prediction uncertainty of the one or more vectors; and configuring one or more stimulation parameters for the recipient device based on the prediction uncertainty.

10. The method of claim 9, wherein the latent space is randomly sampled via a Monte Carlo method.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC111. The method of claim 9, wherein determining the prediction uncertainty of the one or more vectors comprises: determining, based on the one or more vectors, a probability distribution for each of the one or more stimulation parameters, wherein the one or more stimulation parameters are associated with one or more electrodes of the recipient device; determining a variance of the probability distribution exceeds a threshold value; and upon determining the variance exceeds the threshold value, selecting a subset of the one or more electrodes to be configured.

12. The method of claim 11, wherein configuring the one or more stimulation parameters for the recipient device based on the prediction uncertainty comprises: configuring the one or more stimulation parameters associated with the subset of the one or more electrodes.

13. The method of claim 9, 10, 11, or 12, wherein the trained autoencoder network is generated based on training an autoencoder network at one or more user-defined time intervals.

14. The method of claim 9, 10, 11, or 12, wherein the trained autoencoder network includes an encoder and the decoder trained based on an error-based loss function.

15. The method of claim 9, 10, 11, or 12, wherein the one or more stimulation parameters includes observed and unobserved parameters for configuring the recipient device.

16. A system, comprising: a memory; and at least one processor operable coupled to the memory, wherein the at least one processor is configured to: generate, via a decoder of a trained autoencoder network, a plurality of predicted unobserved stimulation parameters for a recipient device; select one or more electrodes of the recipient device to be configured based on the plurality of predicted unobserved stimulation parameters; andAtty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 configure one or more unobserved stimulation parameters of the one or more electrodes.

17. The system of claim 16, wherein the plurality of predicted unobserved stimulation parameters are associated with a time interval during which one or more observed stimulation parameters are missing.

18. The system of claim 16 or 17, wherein select the one or more electrodes of the recipient device to be configured comprises: determine a prediction uncertainty associated with each of the plurality of predicted unobserved stimulation parameters; and select the one or more electrodes based on the prediction uncertainty.

19. The system of claim 16 or 17, wherein one or more weights of the decoder remain fixed while a plurality of latent variables of the decoder are updated.

20. The system of claim 19, wherein the plurality of latent variables is randomly initialized or initialized following a grid.

21. The system of claim 18, wherein the prediction uncertainty is determined based on a probability distribution for each of the plurality of predicted unobserved stimulation parameters.

22. The system of claim 21, wherein the prediction uncertainty is determined based on a variance of the probability distribution.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC123. The system of claim 22, wherein select the one or more electrodes of the recipient device to be configured comprises: select the one or more electrodes based on the variance of the probability distribution associated with the one or more electrodes being greater than a predetermined threshold.

24. The system of claim 16 or 17, wherein recipient device is worn by or implanted in a recipient.

25. One or more non-transitory computer readable storage media comprising instructions that, when executed by a processor, cause the processor to: generate, via a trained machine learning model, a plurality of operational parameters; determine a prediction uncertainty of the plurality of operational parameters; select one or more electrodes of a recipient device based on the prediction uncertainty; and configure the one or more electrodes based on the plurality of operational parameters.

26. The one or more non-transitory computer readable storage media of claim 25, further comprising instructions that, when executed by a processor, cause the processor to: generate, via the trained machine learning model, a plurality of predicted operational parameters of the recipient device based on a latent vector having a plurality of latent variables; and iteratively backpropagate a difference between the plurality of predicted operational parameters and a plurality of observed operational parameters of the recipient device to generate an updated latent vector.

27. The one or more non-transitory computer readable storage media of claim 26, wherein the plurality of operational parameters is generated based on the updated latent vector.

28. The one or more non-transitory computer readable storage media of claim 26, wherein the plurality of latent variables is randomly initialized.Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC129. The one or more non-transitory computer readable storage media of claim 25, 26, 27 or 28, wherein the prediction uncertainty is determined based on a probability distribution for each of the plurality of operational parameters.

30. The one or more non-transitory computer readable storage media of claim 29, wherein the prediction uncertainty is determined based on a mean and a variance of the probability distribution.

31. The one or more non-transitory computer readable storage media of claim 30, wherein the mean and the variance of the probability distribution are displayed to a user via a graphical user interface.

32. The one or more non-transitory computer readable storage media of 25, 26, 27 or 28, wherein the plurality of latent variables is obtained via sampling a latent space.

33. The one or more non-transitory computer readable storage media of claim 32, wherein the latent space is randomly sampled via a Monte Carlo method.

34. The one or more non-transitory computer readable storage media of 25, 26, 27 or 28, wherein the trained machine learning model is obtained by training a machine learning model using an error-based loss function.

35. The one or more non-transitory computer readable storage media of 25, 26, 27 or 28, wherein the plurality of operational parameters comprises one or more of: threshold levels associated with the one or more electrodes, comfort levels associated with the one or more electrodes, or frequency maps associated with the one or more electrodes.

36. A system, comprising: a memory; and at least one processor operable coupled to the memory, wherein the at least one processor is configured to: obtain one or more latent vectors each having a plurality of latent variables;Atty. Docket No. 3065.0865i Client Ref. No. CID04035WOPC1 generate, via a decoder of a trained autoencoder network, one or more predicted vectors each having a plurality of predicted stimulation parameters for a device based the one or more latent vectors; and configure one or more stimulation parameters for the device based on the plurality of predicted stimulation parameters.

37. The system of claim 36, wherein the device is worn by or implanted in a recipient.

38. The system of claim 36 or 37, wherein configure the one or more stimulation parameters for the device based on the plurality of predicted stimulation parameters comprises: configure the one or more stimulation parameters for one or more electrodes of the device based on the plurality of predicted stimulation parameters.

39. The system of claim 38, wherein the one or more stimulation parameters comprises one or more of: threshold levels associated with the one or more electrodes, comfort levels associated with the one or more electrodes, or frequency maps associated with the one or more electrodes.

40. The system of claim 36 or 37, wherein one or more weights of the decoder are fixed.

41. The system of claim 36 or 37, wherein the plurality of latent variables is randomly initialized.

42. The system of claim 36 or 37, wherein the at least one processor is further configured to: determine one or more mean-square errors between the one or more predicted vectors and one or more observed vectors each having a plurality of observed stimulation parameters for the device; and select a subset of the one or more predicted vectors based on the one or more meansquare errors.

Images