System and method for training recurrent neural network for locating devices in environment

Abstract

The present disclosure provides a method and system for training a neural network adapted to locate a device within an environment based on signals received by the device. The method comprises the steps of: training a dual regression device neural network to identify a location from tag data, where the dual regression device neural network comprises a feature extractor and a dual regression device comprising two regression devices; training parameters of double regressors using the marked data and the unmarked data such that each of the two regressors identifies the same marked position while processing the marked data and identifies a different position while processing the unmarked data; and training parameters of the feature extractor using the adversarial discriminator to extract domain invariant features from unlabeled data having statistical properties of the labeled data according to the adversarial discriminator such that each of the two regressors identifies the same location while processing the domain invariant features.

Description

Technical Field

[0001] The present disclosure relates generally to localization of devices and, more particularly, to a system and method for training a recurrent neural network to localize a device in an environment. Background Art

[0002] Domain adaptation (DA) aims to transfer knowledge from a well-labeled source domain to facilitate unlabeled target learning. When turning to specific tasks such as indoor (Wi-Fi) localization, learning cross-domain regressors to mitigate domain shift is critical. However, compared to classification tasks, tasks such as indoor (Wi-Fi) localization rely on regression because the trained neural network needs to recognize locations in a continuous space, while the training data actually only comes from samples at discrete locations. Summary of the invention

[0003] Technical issues

[0004] As the counterpart of classification, regression is the dominant paradigm with a wide range of applications. Domain adaptive regression (DAR) extends the regressor by generalizing it from a labeled source domain to an unlabeled target domain. Deep learning has brought remarkable changes in different regression applications across many fields. However, training high-quality deep models relies on large-scale labeled datasets. Moreover, in many real-world regression applications, it is time-consuming and laborious to accurately annotate a large number of training instances. A solution to this problem is to leverage readily available labeled data from related domains and apply domain adaptation methods to overcome domain shift or dataset bias.

[0005] Deep domain adaptation methods have made significant progress in the domain adaptive classification (DAC) problem. However, despite these advances in DAC, DAR is still not perfect in terms of learning invariant representations in terms of deep mechanisms. One reason for this deficiency lies in the difference in loss functions during the training of classifiers and regressors. The commonly used loss function in regression is the squared loss (L2), while in classification it is the cross entropy loss (CE) with a softmax activation function. Softmax allows activation values ​​of different classes to compete with each other. Introducing this competition mechanism allows the classifier to quickly adapt to changes in feature scales. However, in regression tasks, the regressor may not have this adaptability.

[0006] Therefore, domain adaptation methods for regressors are needed.

[0007] Solution to the problem

[0008] Some embodiments are based on the recognition that the challenges of recurrent neural networks in domain adaptation can be addressed using an adversarial discriminator applied at the feature extraction level. The goal of such a discriminator is to train a feature extractor to extract features from measurements in the target domain that have a similar statistical distribution as the features extracted from measurements in the source domain. However, implementing such a discriminator is problematic because the feature extractor is neither a classifier nor a regressor, and therefore requires additional arbitration to judge the quality of the extracted features. Typical regression architectures for device localization do not have such arbitration.

[0009] To this end, some embodiments disclose a dual regressor neural network and a three-stage training process for domain adaptive regression. The dual regressor neural network includes a feature extractor and a dual regressor including two regressors. The two regressors have the same architecture, but the weights and biases may be different. The three-stage training process is used to train the dual regressor neural network to locate the device in the environment based on the signal received by the device located in the environment. The signal may include different types of radio frequency (RF) signals of various signal levels, such as received signal strength indicator (RSSI) signals, channel state information (CSI) signals. Different types of RF signals include ultra-wideband (UWB) signals, Wi-Fi signals, and inertial measurement unit (IMU) signals. The three-stage training process includes training the dual regressor neural network using labeled data in the environment and / or unlabeled data in a modified environment or a new environment. In an embodiment, the labeled data includes measurements marked with coordinates of the location of the measurement of the signal. In an embodiment, the unlabeled data includes signal measurements at unidentified locations within the environment, so that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment.

[0010] The individual training stages of the three-stage training process are described below. During the first training stage, a dual regressor neural network is trained using labeled data to identify locations in the continuous space of the environment from the labeled data. In particular, a feature extractor is trained to extract features from the labeled data, and each regressor is trained to determine locations in the continuous space of the environment from the features extracted by the feature extractor. In other words, the goal of each regressor is to learn a regression that can map the extracted features to locations in the continuous space of the environment. The need for two regressors comes from the need to arbitrate the features extracted by the feature extractor and the ability to determine whether the input sensor signal is collected from an environment similar to the training data or from an unrecognized environment.

[0011] During the first training phase, based on the loss function L rThe feature extractor and the two regressors are trained using the labeled data. As a result, the parameters of the two regressors are likely to be the same or very similar. Therefore, when both regressors process the features of the labeled data, the two regressors generate similar outputs that approximate the true value data. In an embodiment, the loss function L r Based on the mean squared error (MSE) criterion between labeled data and coordinate estimates.

[0012] During the second training phase, two regressors are trained using labeled data and unlabeled data to generate the same correct output for labeled data, but different outputs for unlabeled data from different domains. In this way, if the statistical distribution of the unlabeled data is different from the statistical distribution of the labeled data, the two regressors are trained to generate different outputs for the unlabeled data. As a result, the two regressors are trained to identify whether the statistical distribution of the unlabeled data has a distribution similar to or different from the statistical distribution of the labeled data. As a result, the two regressors can act as an arbitrator for the similarity between the input statistical distribution and the statistical distribution of the labeled data.

[0013] In addition, the difference in the outputs of the two regressors for unlabeled data can be maximized by minimizing the overlap / similarity between the outputs of the two regressors. In order to minimize the overlap between the outputs of the two regressors, the overlap between the outputs of the two regressors needs to be quantified. To quantify this overlap, some embodiments use the Jaccard similarity coefficient (IoU score) from the field of object detection using visual sensors. However, the non-differentiability of the IoU score makes the optimization of the weights of the dual regressor neural network problematic. To alleviate this problem, a soft similarity function L is implemented. s Soft similarity function L s Minimizing reduces the overlap between the outputs of the two regressors to further increase the output difference of the two regressors for unlabeled data.

[0014] However, it is still required that both regressors should generate the same output for the labeled data. To this end, in the second training phase, instead of only using the soft similarity function L s Minimize, based on the loss function L r The objective function and soft similarity function L (used in the first training stage) s is minimized.

[0015] Therefore, the combination of the two regressors can be viewed as an implicit discriminator to directly distinguish labeled data from unlabeled data (i.e., detect target samples outside of source samples / support).

[0016] Taking into account these new capabilities of the two regressors, during the third training phase, a feature extractor is trained using an adversarial discriminator to extract domain-invariant features of unlabeled data with the statistical distribution of labeled data using the two regressors as arbitration. The extracted domain-invariant features enable each of the two regressors to identify the same location while processing the domain-invariant features when processed by the two regressors.

[0017] Some embodiments are based on the recognition that domain adaptation becomes difficult if the distribution difference is large. As used herein, distribution difference refers to the difference in the outputs of two regressors. For example, in the case of multimodal indoor positioning, the distribution difference becomes relatively significant because the change of one multipath component may affect the entire sensor signal in a constructive or destructive manner. Some embodiments are based on the recognition that this problem can be solved by constructing two intermediate domains between the source domain and the target domain and gradually eliminating their mismatches to achieve statistical distribution alignment. The intermediate domain is constructed based on data augmentation of labeled data and unlabeled data.

[0018] First, a fixed ratio λ is selected, for example, λ is set to 0.7. Based on the fixed λ, the labeled data and the unlabeled data are linearly combined to synthesize the enhanced labeled data and augmenting unlabeled data and Two intermediate domains are formed, in which Closer to the source domain, Closer to the target domain. In addition, the adversarial relationship between the discriminator and the feature extractor is exploited to alleviate the problem of augmenting labeled data. and augmenting unlabeled data For example, augmenting labeled data and augmenting unlabeled data is input to the feature extractor, and the corresponding output is processed by each of the two regressors in the second stage. In addition, the adversarial loss function is minimized to mitigate the domain shift between the source similar domain and the target similar domain.

[0019] After performing the above three-stage training process, a trained dual regressor neural network is obtained. The trained dual regressor neural network includes a trained feature extractor and two trained regressors. The trained dual regressor neural network can be used to determine the position for the input unlabeled data. Here, the two trained regressors of the trained dual regressor neural network generate similar positions for the input unlabeled data. To this end, some embodiments are based on the recognition that the architecture of the trained dual regressor neural network may include only a single trained regressor, rather than two trained regressors, for determining the position. Alternatively, in some embodiments, the trained dual regressor neural network may include a trained feature extractor and an original regressor (a regressor obtained from the first training stage), which may provide additional robustness of the regression output.

[0020] Thus, one embodiment discloses a computer-implemented method for training a neural network suitable for locating a device within an environment based on signals received by the device located in the environment. The method uses a processor using stored instructions implementing the method, wherein the instructions, when executed by the processor, perform the steps of the method, the method comprising the steps of: collecting labeled data comprising measurements labeled with coordinates of locations of the measurements of the signals; collecting unlabeled data comprising measurements of the signals at unidentified locations within the environment, such that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment; during a first training phase, training a dual regressor neural network using the labeled data to identify locations in a continuous space of the environment from the labeled data, wherein the dual regressor neural network comprises a feature extractor and a dual regressor, the feature extractor being configured to extract features from the labeled data, the dual regressor comprising two regressors having the same architecture, each regressor being trained to determine a position in a continuous space of an environment from features received from a feature extractor; during a second training stage, using fixed parameters of the feature extractor trained during the first training stage, training parameters of a dual regressor using labeled data and unlabeled data, such that each of the two regressors recognizes the same labeled position while processing the labeled data and recognizes different positions while processing the unlabeled data; and during a third training stage, using the fixed parameters of the dual regressor trained using labeled data and unlabeled data during the second training stage, training parameters of the feature extractor using an adversarial discriminator to extract domain-invariant features from the unlabeled data having statistical properties of the labeled data according to the adversarial discriminator, such that each of the two regressors recognizes the same position while processing the domain-invariant features.

[0021] Therefore, another embodiment discloses a system for training a neural network suitable for locating a device within an environment based on a signal received by the device located in the environment. The system includes a processor and a memory storing instructions, which when executed by the processor causes the system to: collect labeled data, which includes measurements marked with coordinates of the location of the measurement of the signal; collect unlabeled data, which includes signal measurements at unidentified locations within the environment, so that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment; during a first training phase, use the labeled data to train a dual regressor neural network to identify locations in a continuous space of the environment from the labeled data, wherein the dual regressor neural network includes a feature extractor and a dual regressor, the feature extractor is configured to extract features from the labeled data, and the dual regressor includes two regressors having the same architecture, each regressor is trained to extract features from the features received from the feature extractor. features to determine a position in a continuous space of an environment; during a second training stage, using fixed parameters of the feature extractor trained during the first training stage, training parameters of a dual regressor using labeled data and unlabeled data, such that each of the two regressors identifies the same labeled position while processing the labeled data and identifies different positions while processing the unlabeled data; and during a third training stage, using the fixed parameters of the dual regressor trained using labeled data and unlabeled data during the second training stage, training parameters of the feature extractor using an adversarial discriminator to extract domain invariant features from the unlabeled data having statistical properties of the labeled data according to the adversarial discriminator, such that each of the two regressors identifies the same position while processing the domain invariant features.

[0022] Therefore, another embodiment discloses a non-transitory computer-readable storage medium embodied with a program, which can be executed by a processor to perform a method for training a neural network, which is suitable for locating a device within an environment based on a signal received by the device located in the environment. The method includes the following steps: collecting labeled data, which includes measurements marked with coordinates of the location of the measurement of the signal; collecting unlabeled data, which includes signal measurements at unidentified locations within the environment, so that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment; during a first training phase, using the labeled data to train a dual regressor neural network to identify locations in a continuous space of the environment from the labeled data, wherein the dual regressor neural network includes a feature extractor and a dual regressor, the feature extractor is configured to extract features from the labeled data, and the dual regressor includes two regressors with the same architecture, each regressor is trained to determine the location in the continuous space of the environment from the features received from the feature extractor. position; during a second training stage, using the fixed parameters of the feature extractor trained during the first training stage, training the parameters of the dual regressor using the labeled data and the unlabeled data, so that each of the two regressors recognizes the same labeled position while processing the labeled data and recognizes different positions while processing the unlabeled data; and during a third training stage, using the fixed parameters of the dual regressor trained using the labeled data and the unlabeled data during the second training stage, training the parameters of the feature extractor using the adversarial discriminator to extract domain invariant features from the unlabeled data having the statistical properties of the labeled data according to the adversarial discriminator, so that each of the two regressors recognizes the same position while processing the domain invariant features.

[0023] Embodiments of the present disclosure will be further described with reference to the accompanying drawings. The drawings shown are not necessarily to scale, emphasis generally being placed upon illustrating the principles of embodiments of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] [ Figure 1A ] Figure 1A A schematic diagram showing the architecture of a dual regressor neural network according to some embodiments of the present disclosure.

[0025] [ Figure 1B ] Figure 1B A schematic diagram illustrating a second training phase according to some embodiments of the present disclosure.

[0026] [ Figure 1C ] Figure 1C A schematic diagram illustrating a third training phase according to some embodiments of the present disclosure.

[0027] [ Figure 2A ] Figure 2AA schematic diagram showing the architecture of a trained dual regressor neural network for determining the location of a device for input unlabeled data according to some embodiments of the present disclosure.

[0028] [ Figure 2B ] Figure 2B The architecture of a dual regressor neural network for determining a location of a device according to some embodiments of the present disclosure is shown, wherein the dual regressor neural network includes a single trained regressor.

[0029] [ Figure 2C ] Figure 2C The architecture of a dual regressor neural network for determining the location of a device according to some other embodiments of the present disclosure is shown.

[0030] [ Figure 3 ] Figure 3 The control of a robot in an automated factory according to some embodiments of the present disclosure is illustrated.

[0031] [ Figure 4 ] Figure 4 is a schematic diagram illustrating a system for training a neural network suitable for locating a device within an environment according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0032] In the following description, for the purpose of illustration, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be apparent to those skilled in the art that the present disclosure may be practiced without these specific details. In other cases, the devices and methods are shown only in block diagram form to avoid obscuring the present disclosure.

[0033] As used in this specification and claims, the terms "for example" and "such as" and the verbs "include," "have," "comprises," and other verb forms thereof, when used in conjunction with a list of one or more components or other items, should each be interpreted as open-ended, meaning that the list should not be viewed as excluding other additional components or items. The term "based on" means based at least in part on. In addition, it will be understood that the phraseology and terminology employed herein are for descriptive purposes and should not be viewed as limiting. Any headings utilized within this description are for convenience only and have no legal or limiting effect.

[0034] Domain adaptation (DA) aims to transfer knowledge from a well-labeled source domain to facilitate unlabeled target learning. When turning to specific tasks such as indoor (Wi-Fi) localization, learning cross-domain regressors to mitigate domain shift is critical. However, compared to classification tasks, tasks such as indoor (Wi-Fi) localization rely on regression because the trained neural network needs to recognize locations in a continuous space, while the training data actually only comes from samples at discrete locations.

[0035] Regression is a technique used to determine the relationship between independent variables or features and a dependent variable or outcome. Regression can be used as a predictive modeling method in machine learning, where algorithms are used to predict continuous outcomes. Domain adaptive regression extends regression by generalizing the regressor from a labeled source domain to an unlabeled target domain.

[0036] For example, a regressor may be trained with labeled data in one environment. However, during the online testing phase, the regressor may receive unlabeled data in a new environment. As a result, a regressor trained with labeled data may suffer from performance degradation when the regressor processes unlabeled data. For this reason, domain adaptation methods for regressors are needed.

[0037] Some embodiments are based on the recognition that the challenges of recurrent neural networks in domain adaptation can be addressed using an adversarial discriminator applied at the feature extraction level. The goal of such a discriminator is to train a feature extractor to extract features from measurements in the target domain that have a similar statistical distribution as the features extracted from measurements in the source domain. However, implementing such a discriminator is problematic because the feature extractor is neither a classifier nor a regressor, and therefore requires additional arbitration to judge the quality of the extracted features. Typical regression architectures for device localization do not have such arbitration.

[0038] To this end, some embodiments disclose a dual-regressor neural network and a three-stage training process for domain adaptive regression. Figure 1A A schematic diagram showing the architecture of a dual regressor neural network 100 according to some embodiments of the present disclosure. The dual regressor neural network 100 includes a feature extractor F(·) 103 and a dual regressor including two regressors 105 and 107. and 105 and 107 respectively. The two regressors 105 and 107 have the same architecture. The three-stage training process is used to train the dual regressor neural network 100 to locate the device in the environment based on the signal received by the device in the environment. The signal may include different types of radio frequency (RF) signals of various signal levels, such as received signal strength indicator (RSSI) signals, channel state information (CSI) signals. Different types of RF signals include ultra-wideband (UWB) signals, Wi-Fi signals, and inertial measurement unit (IMU) signals. The three-stage training process includes training the dual regressor neural network 100 using labeled data 101 and / or unlabeled data. In an embodiment, the labeled data 101 includes measurements marked with coordinates of the location of the measurement of the signal. In an embodiment, the unlabeled data includes signal measurements at unidentified locations within the environment, so that the measurements of the labeled data 101 and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment.

[0039] The individual training stages of the three-stage training process are described below.

[0040] During the first training phase, the dual regressor neural network 100 is trained using labeled data 101 to identify locations in the continuous space of the environment from the labeled data 101. In particular, the feature extractor 103 is trained to extract features from the labeled data 101, and each regressor (i.e., 105 and 107) is trained to determine the location in the continuous space of the environment from the features extracted by the feature extractor 103. In other words, the goal of each regressor is to learn a regression that can map the extracted features to locations in the continuous space of the environment. The need for two regressors 105 and 107 comes from the need to arbitrate the features extracted by the feature extractor 103.

[0041] In some embodiments, during a first training phase, the feature extractor 101 and the two regressors 105 and 107 are trained using the labeled data 101 based on a loss function Lr. As a result, the parameters of the two regressors 105 and 107 are mostly similar. Therefore, when the two regressors 105 and 107 process the extracted features, the two regressors 105 and 107 generate similar labeled positions that approximate the true value data. In an embodiment, the loss function Lr is based on a mean square error (MSE) criterion. For example, the first training phase is performed in a supervised manner.

[0042]

[0043] in is the MSE loss, is the true value data, is the output of regressor 105, is the output of regressor 107.

[0044] Figure 1B A schematic diagram of a second training phase according to some embodiments of the present disclosure is shown. During the second training phase, the parameters of the feature extractor 101 trained during the first training phase are fixed, and the two regressors 105 and 107 are trained using the labeled data 101 and the unlabeled data 109 to generate the same correct labeled position for the labeled data 101, but different positions for the unlabeled data 109. In this way, if the statistical distribution of the unlabeled data 109 is different from the statistical distribution of the labeled data 101, the two regressors 105 and 107 are trained to generate different outputs (i.e., positions) for the unlabeled data 109. As a result, the two regressors 105 and 107 are trained to identify whether the statistical distribution of the unlabeled data 109 has a distribution similar to or different from the statistical distribution of the labeled data 101. As a result, the two regressors 105 and 107 can act as an arbitration of the statistical distribution.

[0045] In addition, the difference in the outputs of the two regressors 105 and 107 for unlabeled data can be maximized by minimizing the overlap / similarity between the two regressors 105 and 107. In order to minimize the overlap between the two regressors 105 and 107, the overlap between the outputs of the two regressors needs to be quantified. In order to quantify the overlap between the two regressors 105 and 107, some embodiments use the Jaccard similarity coefficient (IoU score) from the field of object detection using visual sensors. However, the non-differentiability of the IoU score makes the optimization of the weights of the dual regressor neural network 100 problematic. In order to alleviate this problem, a soft similarity function Ls 111 between the cascaded outputs of the two components G and R of the two regressors is implemented. The output of the first regression component is denoted as G(f), and the output of the second regression component is denoted as R(G(f)) (also the output of regression). The cascade output h = [G, R(G(f))]. For unlabeled data, at the intermediate and final outputs of the two regressors,

[0046]

[0047] Minimization of the soft similarity function Ls 111 reduces the overlap between the two regressors 105 and 107 to further increase the output difference of the two regressors 105 and 107 for the unlabeled data 109. In an embodiment, the soft similarity function Ls 111 is given as

[0048]

[0049] Among them, <·,·> indicates and The inner product of the intersection of is the element-wise product, and d is The dimension size of .

[0050] However, it is still required that both regressors 105 and 107 should generate the same output for the labeled data 101. To this end, in the second training phase, instead of only using the soft similarity function L s 111 Minimization, based on the loss function L r The objective function and soft similarity function L (used in the first training stage) s 111 is minimized. The objective function to be minimized is given as

[0051]

[0052] Therefore, the combination of the two regressors 105 and 107 can be regarded as an implicit discriminator to directly distinguish the labeled data 101 from the unlabeled data 109.

[0053] Figure 1CA schematic diagram of the third training stage according to some embodiments of the present disclosure is shown. In the third training stage, the parameters of the feature extractor 103 are trained using the adversarial discriminator 113 using the fixed parameters of the two regressors 105 and 107 trained using the labeled data and the unlabeled data during the second training stage. During the third training stage, the parameters of the feature extractor 103 are trained using the adversarial discriminator 113 to extract domain-invariant features of the unlabeled data 109 having the statistical distribution of the labeled data 101 using the two regressors 105 and 107 as arbitration. The extracted domain-invariant features enable each of the two regressors 105 and 107 to identify the same location while processing the domain-invariant features when processed by the two regressors 105 and 107.

[0054] In addition, some embodiments are based on the recognition that domain adaptation becomes difficult if the distribution difference is large. As used herein, distribution difference refers to the difference in the outputs of the two regressors 105 and 107. For example, in the case of multimodal indoor positioning, the distribution difference becomes relatively significant. Some embodiments are based on the recognition that this problem can be solved by constructing two intermediate domains between the source domain and the target domain and gradually eliminating their mismatches to achieve statistical distribution alignment. The intermediate domain is constructed based on data enhancement of the labeled data 101 and the unlabeled data 109. First, a fixed ratio λ is selected, for example, λ is set to 0.7. Based on the fixed λ, the labeled data and the unlabeled data are linearly combined to synthesize the enhanced labeled data 115 and augmenting unlabeled data 117:

[0055]

[0056] and Two intermediate domains are formed, where the enhanced labeled data 115 Similar to the source domain (or labeled data 105), enhance the unlabeled data 117 is similar to the target domain (or unlabeled data 107). In addition, the adversarial relationship between the adversarial discriminator 113 and the feature extractor 103 is used to mitigate the enhancement of the labeled data. 115 and augmenting unlabeled data 117. For example, enhancing labeled data 115 and augmenting unlabeled data 117 is input to the feature extractor 103, and the corresponding output is processed by each of the two regressors 105 and 107 in the second stage to output and

[0057] In addition, the adversarial loss function is adopted

[0058]

[0059] in represents the source similarity or target similarity domain, and D(·) is the adversarial discriminator 113. To this end, mathematically, the third training stage is given as:

[0060]

[0061] After performing the above three-stage training process, a trained dual regressor neural network is obtained. During the online (ie, real-time) stage, the trained dual regressor neural network can be used to determine the location of the device in the environment for input unlabeled data.

[0062] Figure 2A A schematic diagram showing the architecture of a trained dual regressor neural network 200 for determining the location of a device for input unlabeled data 201 according to some embodiments of the present disclosure. The trained dual regressor neural network 200 includes a trained feature extractor 203 obtained from the third training stage and two trained regressors 205 and 207 obtained from the second training stage. The unlabeled data 201 is input to the trained feature extractor 203. The trained feature extractor 203 extracts domain-invariant features from the unlabeled data 201. In addition, the extracted domain-invariant features are processed by the two trained regressors 205 and 207. Each of the two trained regressors 205 and 207 outputs the same location of the device in the environment.

[0063] Since the two trained regressors 205 and 207 of the trained dual regressor neural network 200 output the same position for the input unlabeled data 201, some embodiments are based on the recognition that the architecture of the trained dual regressor neural network 200 may include only one trained regressor, rather than two trained regressors, for determining the position. Figure 2B This architecture of a dual regressor neural network is shown in .

[0064] Figure 2B The architecture of a dual regressor neural network 209 for determining a location according to some other embodiments of the present disclosure is shown. The dual regressor neural network 209 includes a trained feature extractor 203 and a trained regressor 207. Unlabeled data 201 is input to the trained feature extractor 203. The trained feature extractor 203 extracts domain-invariant features from the unlabeled data 201. In addition, the extracted domain-invariant features are processed by the trained regressor 207 to output the location of the device in the environment.

[0065] Alternatively, in some implementations, dual regressor neural network 209 may include trained regressor 205 instead of trained regressor 207 .

[0066] Some embodiments are based on the recognition that a trained dual regressor neural network for determining position may include a trained feature extractor 203 and an original regressor (the regressor obtained from the first training phase). Figure 2C Such a dual regressor neural network is shown in . Figure 2C The architecture of a dual regressor neural network 211 for determining a position according to some other alternative embodiments of the present disclosure is shown. The dual regressor neural network 211 includes the trained feature extractor 203 obtained from the third training stage and the two regressors 105 and 107 obtained from the first training stage. Alternatively, in some embodiments, the dual regressor neural network 211 may include the trained feature extractor 203 obtained from the third training stage and one of the two regressors 105 and 107 obtained from the first training stage.

[0067] According to some embodiments, the position of a device in an environment output by a trained dual regressor neural network (e.g., trained dual regressor neural network 200 or 209) may be used to control the device in the environment to perform a task. For example, the device may be a robot and the environment may be an automated factory. Based on the position of the robot in the automated factory, the robot may be controlled to perform a task, such as reaching a target position, as described below. Figure 3 Described in .

[0068] Figure 3 The control of a robot 301 in an automated factory 300 according to some embodiments of the present disclosure is illustrated. The robot 301 is embedded with a trained dual regressor neural network 200. In addition, the robot 301 includes a controller 303. The robot 301 receives unlabeled data, which includes signal measurements at unidentified locations within the automated factory 300. The unlabeled data is input to the trained dual regressor neural network 200 to determine a location 305 of the robot 301 in the automated factory 300. Based on the location 305 of the robot 301, the controller 303 determines a motion path 307 connecting the location 303 in the automated factory 300 and a target location 309. In addition, the controller 303 controls the robot 301 according to the motion path 307. In some alternative embodiments, the controller 303 may make decisions about obstacle avoidance based on the location 305 of the robot 301.

[0069] Additionally, in some embodiments, the robot 301 may also be tracked in the automated factory 300 based on the location of the robot 301. Additionally or alternatively, the trained dual regressor neural network 200 may be used in any environment where asset (e.g., robot) tracking is necessary or critical, such as an automated factory 300, a warehouse, or a smart home.

[0070] Figure 44 is a schematic diagram illustrating a system 400 for implementing the method of the present disclosure. The computing device 400 includes a power supply 401, a processor 403, a memory 405, a storage device 407, all of which are connected to a bus 409. In addition, a high-speed interface 411, a low-speed interface 413, a high-speed expansion port 415, and a low-speed connection port 417 can be connected to the bus 409. In addition, a low-speed expansion port 419 is connected to the bus 409. In addition, an input interface 421 can be connected to an external receiver 423 and an output interface 425 via the bus 409. The receiver 427 can be connected to an external transmitter 429 and a transmitter 431 via the bus 409. An external memory 433, an external sensor 435, a machine 437, and an environment 439 can also be connected to the bus 409. In addition, one or more external input / output devices 441 can be connected to the bus 409. Network interface controller (NIC) 443 may be adapted to connect to network 445 via bus 409 , wherein data or other data may be rendered on a third-party display device, third-party imaging device, and / or third-party printing device external to computing device 400 .

[0071] The memory 405 can store instructions executable by the computing device 400 and any data that can be utilized by the methods and systems of the present disclosure. The memory 405 can include random access memory (RAM), read-only memory (ROM), flash memory, or any other suitable memory system. The memory 405 can be a volatile memory unit and / or a non-volatile memory unit. The memory 405 can also be another form of computer-readable medium, such as a magnetic disk or an optical disk.

[0072] The storage device 407 may be adapted to store supplemental data and / or software modules used by the computer device 400. The storage device 407 may include a hard drive, an optical drive, a thumb drive, a drive array, or any combination thereof. In addition, the storage device 407 may include a computer readable medium such as a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory or other similar solid-state memory device, or an array of devices including devices in a storage area network or other configuration. Instructions may be stored in an information carrier. The instructions, when executed by one or more processing devices (e.g., processor 403), perform one or more methods such as those described above.

[0073] Optionally, the computing device 400 may be linked via the bus 409 to a display interface or user interface (HMI) 447 adapted to connect the computing device 400 to a display device 449 and a keyboard 451, wherein the display device 449 may include a computer monitor, a camera, a television, a projector, or a mobile device, etc. In some implementations, the computer device 400 may include a printer interface to connect to a printing device, wherein the printing device may include a liquid inkjet printer, a solid ink printer, a large commercial printer, a thermal printer, a UV printer, or a dye sublimation printer, etc.

[0074] The high-speed interface 411 manages the broadband intensive operations of the computing device 400, while the low-speed interface 413 manages the lower bandwidth intensive operations. This functional allocation is only an example. In some implementations, the high-speed interface 411 can be connected to the memory 405, the user interface (HMI) 444, and to the keyboard 451 and the display 449 (e.g., through a graphics processor or accelerator), and to the high-speed expansion port 415, which can accept various expansion cards via the bus 409.

[0075] In an implementation, the low-speed interface 413 is coupled to the storage device 407 and the low-speed expansion port 417 via the bus 409. The low-speed expansion port 417, which may include various communication ports (e.g., USB, Bluetooth, Ethernet, wireless Ethernet), may be coupled to one or more input / output devices 441. The computing device 400 may be connected to the server 453 and the rack server 455. The computing device 400 may be implemented in a variety of different forms. For example, the computing device 400 may be implemented as part of the rack server 455.

[0076] The above description provides only exemplary embodiments and is not intended to limit the scope, applicability or configuration of the present disclosure. On the contrary, the following description of the exemplary embodiments will provide a feasible description for implementing one or more exemplary embodiments to those skilled in the art. It is contemplated that various changes may be made to the functions and arrangements of the elements without departing from the spirit and scope of the disclosed subject matter set forth in the appended claims.

[0077] Specific details are given in the following description to provide a thorough understanding of the embodiments. However, it will be appreciated by those of ordinary skill in the art that the embodiments may be practiced without these specific details. For example, the systems, processes, and other elements in the disclosed subject matter may be shown as components in block diagram form to avoid obscuring the embodiments in terms of unnecessary details. In other cases, well-known processes, structures, and techniques may be shown without unnecessary details to avoid obscuring the embodiments. In addition, similar reference numerals and designations in the various drawings indicate similar elements.

[0078] In addition, various embodiments may be described as processes, which are depicted as flow charts, data flow diagrams, structure diagrams, or block diagrams. Although flow charts may describe operations as sequential processes, many operations may be performed in parallel or simultaneously. In addition, the order of operations may be rearranged. A process may terminate when its operations are completed, but may have additional steps that are not discussed or included in the accompanying drawings. In addition, not all operations in any specifically described process may appear in all embodiments. A process may correspond to a method, a function, a program, a subroutine, a subprogram, etc. When a process corresponds to a function, the termination of the function may correspond to the function returning to the calling function or the main function.

[0079] In addition, embodiments of the disclosed subject matter may be implemented at least in part manually or automatically. Manual or automatic implementation may be performed or at least assisted by the use of a machine, hardware, software, firmware, middleware, microcode, hardware description language, or any combination thereof. When implemented in software, firmware, middleware, or microcode, program code or code segments that perform the desired tasks may be stored in a machine-readable medium. A processor may perform the desired tasks.

[0080] The various methods or processes outlined herein may be encoded as software that can be executed on one or more processors using any of a variety of operating systems or platforms. In addition, such software may be written using any of a variety of suitable programming languages ​​and / or programming or scripting tools, and may also be compiled into executable machine language code or intermediate code that is executed on a framework or virtual machine. Typically, in various embodiments, the functionality of the program modules may be combined or distributed as desired.

[0081] The embodiments of the present disclosure may be embodied as a method, examples of which have been provided. The actions performed as part of the method may be ordered in any suitable manner. Thus, embodiments may be constructed that perform the actions in a different order than shown, which may include performing some actions simultaneously, although shown as sequential actions in the illustrative embodiments.

[0082] In addition, the embodiments of the present disclosure and the functional operations described in this specification may be implemented in digital electronic circuits, in tangibly implemented computer software or firmware, in computer hardware including the structures disclosed in this specification and their structural equivalents, or in a combination of one or more of them. In addition, some embodiments of the present disclosure may be implemented as one or more computer programs, that is, one or more modules of computer program instructions that are implemented on a tangible non-transitory program carrier for execution by a data processing device or control the operation of the data processing device. In addition, the program instructions may be encoded on an artificially generated propagation signal, such as a machine-generated electrical signal, optical signal, or electromagnetic signal, which is generated to encode information for transmission to a suitable receiver device for execution by a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0083] According to an embodiment of the present disclosure, the term "data processing device" may cover all types of devices, apparatuses and machines for processing data, including, by way of example, a programmable processor, a computer or a plurality of processors or computers. The device may include dedicated logic circuits, such as an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit). In addition to hardware, the device may also include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.

[0084] A computer program (which may also be referred to or described as a program, software, software application, module, software module, script, or code) may be written in any form of programming language, including compiled or interpreted languages ​​or declarative or procedural languages, and it may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program may (but need not) correspond to a file in a file system. A program may be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files, such as files that store one or more modules, subroutines, or portions of code.

[0085] A computer program may be deployed to be executed on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communication network. As an example, a computer suitable for executing a computer program includes a central processing unit that may be based on a general or special purpose microprocessor, or both, or any other type. Typically, the central processing unit will receive instructions and data from a read-only memory or a random access memory, or both. The basic elements of a computer are a central processing unit for executing instructions and one or more memory devices for storing instructions and data.

[0086] Typically, a computer will also include one or more mass storage devices (e.g., magnetic, magneto-optical disks, or optical disks) for storing data or be operatively connected to receive data from or transfer data to, or both, one or more mass storage devices (e.g., magnetic, magneto-optical disks, or optical disks) for storing data. However, a computer need not have these devices. In addition, a computer may be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, etc.

[0087] To provide interaction with a user, embodiments of the subject matter described in this specification may be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices may also be used to provide interaction with the user; for example, feedback provided to the user may be any form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback; and input from the user may be received in any form, including acoustic, verbal, or tactile input. In addition, a computer may interact with a user by sending documents to and receiving documents from a device used by the user; for example, by sending a web page to a web browser in response to a request received from a web browser on a user's client device.

[0088] Implementations of the subject matter described in this specification may be implemented in a computing system that includes a back-end component (e.g., as a data server), or includes a middleware component (e.g., an application server), or includes a front-end component (e.g., a client computer with a graphical user interface or a web browser through which a user can interact with an implementation of the subject matter described in this specification), or any combination of one or more such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LANs") and wide area networks ("WANs"), such as the Internet.

[0089] A computing system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0090] Although the present disclosure has been described with reference to certain preferred embodiments, it will be understood that various other adaptations and modifications may be made within the spirit and scope of the present disclosure. Therefore, the aspects of the appended claims cover all such changes and modifications that fall within the true spirit and scope of the present disclosure.

Claims

1. A computer-implemented method for training a neural network adapted to locate a device within an environment based on signals received by the device within the environment, wherein: The method uses a processor using stored instructions for implementing the method, wherein the instructions, when executed by the processor, perform the steps of the method, the method comprising the steps of: collecting tagged data comprising said measurements tagged with coordinates of locations of said measurements of said signals; collecting unlabeled data comprising measurements of the signal at unidentified locations within the environment, such that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment; During a first training phase, training a dual regressor neural network using the labeled data to identify locations in a continuous space of the environment from the labeled data, wherein the dual regressor neural network comprises a feature extractor and a dual regressor, the feature extractor being configured to extract features from the labeled data, the dual regressor comprising two regressors having the same architecture, each regressor being trained to determine locations in the continuous space of the environment from features received from the feature extractor; During a second training phase, training the parameters of the dual regressor using the labeled data and the unlabeled data using the fixed parameters of the feature extractor trained during the first training phase, such that each of the two regressors recognizes the same labeled locations while processing the labeled data and recognizes different locations while processing the unlabeled data; and During a third training stage, parameters of the feature extractor are trained using an adversarial discriminator using the fixed parameters of the dual regressor trained using the labeled data and the unlabeled data during the second training stage to extract domain-invariant features from the unlabeled data having statistical properties of the labeled data according to the adversarial discriminator, so that each of the two regressors recognizes the same location while processing the domain-invariant features.

2. The computer-implemented method of claim 1 , wherein: During the first training phase, the dual regressor neural network is trained using a loss function such that each of the two regressors identifies the same location while processing the labeled data, and wherein the loss function is based on a mean squared error (MSE) criterion.

3. The computer-implemented method of claim 1 , wherein: During the second training phase, the parameters of the dual regressor are trained based on a soft similarity function using the fixed parameters of the feature extractor trained during the first training phase, so that each of the two regressors identifies different locations while processing the unlabeled data.

4. The computer-implemented method of claim 3, wherein: During the second training stage, the parameters of the dual regressor are trained based on an objective function using the fixed parameters of the feature extractor trained during the first training stage, so that each of the two regressors recognizes the same labeled location while processing the labeled data and recognizes different locations while processing the unlabeled data, and wherein the objective function is based on a loss function and the soft similarity function.

5. The computer-implemented method of claim 3, wherein: During the third training stage, parameters of the adversarial discriminator and the feature extractor are trained based on an adversarial loss function and the soft similarity function using fixed parameters of the dual regressor trained during the second training stage, and wherein the adversarial loss function is configured to mitigate domain distribution shift.

6. The computer-implemented method of claim 1 , wherein: During the third training stage, parameters of the adversarial discriminator and the feature extractor are trained based on maximization of an adversarial loss function using fixed parameters of the dual regressor trained during the second training stage.

7. The computer-implemented method of claim 1 , wherein: The method further comprises the step of determining enhanced labeled data and enhanced unlabeled data forming more than one intermediate domain.

8. The computer-implemented method of claim 7, wherein: The method further includes the step of mitigating a domain shift between the enhanced labeled data and the enhanced unlabeled data based on an adversarial relationship between the adversarial discriminator and the feature extractor to align statistical properties of the labeled data and the unlabeled data.

9. The computer-implemented method of claim 7, wherein: The enhanced labeled data is determined by linearly combining the labeled data and the unlabeled data at a fixed ratio so that the enhanced labeled data is similar to the labeled data, and wherein the enhanced unlabeled data is determined by linearly combining the unlabeled data and the labeled data at the fixed ratio so that the enhanced unlabeled data is similar to the unlabeled data.

10. The computer-implemented method of claim 1, wherein: The method further comprises the following steps: Inputting the unlabeled data collected by the device into a trained dual-regressor neural network, wherein the trained dual-regressor neural network includes the feature extractor trained during the third training stage and at least one of the two regressors trained during the second training stage; and The unlabeled data is processed using the trained dual regressor neural network to output the location of the device in the environment.

11. The computer-implemented method of claim 1 , wherein: The method further comprises the following steps: Inputting the unlabeled data collected by the apparatus into a trained dual-regressor neural network, wherein the trained dual-regressor neural network includes the feature extractor trained during the third training phase and at least one of the two regressors trained using the labeled data during the first training phase; and The unlabeled data is processed using the trained dual regressor neural network to output the location of the device in the environment.

12. The computer-implemented method of claim 11, wherein: The method further comprises the following steps: determining a motion path of the device to perform a task in the environment based on the position of the device in the environment; and The device is controlled based on the motion path.

13. The computer-implemented method of claim 11 , wherein: The method further comprises the step of tracking the device in the environment based on the outputted position of the device.

14. The computer-implemented method of claim 11, wherein: The device is a robot, and wherein the environment is one of an automated factory, a warehouse, or a smart home.

15. A system for training a neural network, the neural network being adapted to locate a device within an environment based on signals received by the device within the environment, the system comprising: processor; and a memory storing instructions that, when executed by the processor, cause the system to: collecting tagged data comprising said measurements tagged with coordinates of locations of said measurements of said signals; collecting unlabeled data comprising measurements of the signal at unidentified locations within the environment, such that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment; During a first training phase, training a dual regressor neural network using the labeled data to identify locations in a continuous space of the environment from the labeled data, wherein the dual regressor neural network comprises a feature extractor and a dual regressor, the feature extractor being configured to extract features from the labeled data, the dual regressor comprising two regressors having the same architecture, each regressor being trained to determine locations in the continuous space of the environment from features received from the feature extractor; During a second training phase, the parameters of the dual regressor are trained using the labeled data and the unlabeled data using the fixed parameters of the feature extractor trained during the first training phase, so that each of the two regressors recognizes the same labeled locations while processing the labeled data and recognizes different locations while processing the unlabeled data; and During a third training stage, parameters of the feature extractor are trained using an adversarial discriminator using the fixed parameters of the dual regressor trained using the labeled data and the unlabeled data during the second training stage to extract domain-invariant features from the unlabeled data having statistical properties of the labeled data according to the adversarial discriminator, so that each of the two regressors recognizes the same location while processing the domain-invariant features.

16. The system of claim 15, wherein: During the first training phase, the dual regressor neural network is trained using a loss function such that each of the two regressors identifies the same location while processing the labeled data, and wherein the loss function is based on a mean squared error (MSE) criterion.

17. The system of claim 15, wherein: During the second training phase, the parameters of the dual regressor are trained based on a soft similarity function using the fixed parameters of the feature extractor trained during the first training phase, so that each of the two regressors identifies different locations while processing the unlabeled data.

18. The system of claim 17, wherein: During the second training stage, the parameters of the dual regressor are trained based on an objective function using the fixed parameters of the feature extractor trained during the first training stage, so that each of the two regressors recognizes the same labeled location while processing the labeled data and recognizes different locations while processing the unlabeled data, and wherein the objective function is based on a loss function and the soft similarity function.

19. The system of claim 15, wherein: The processor is further configured to: Inputting the unlabeled data collected by the device into a trained dual-regressor neural network, wherein the trained dual-regressor neural network includes the feature extractor trained during the third training stage and at least one of the two regressors trained during the second training stage; and The unlabeled data is processed using the trained dual regressor neural network to output the location of the device in the environment.

20. A non-transitory computer-readable storage medium having a program implemented thereon, the program being executable by a processor for performing a method for training a neural network, the neural network being adapted to locate a device within an environment based on a signal received by the device within the environment, the method comprising the steps of: collecting tagged data comprising said measurements tagged with coordinates of locations of said measurements of said signals; collecting unlabeled data comprising measurements of the signal at unidentified locations within the environment, such that the measurements of the labeled data and the measurements of the unlabeled data are sampled at the same or different discrete locations of the environment; During a first training phase, training a dual regressor neural network using the labeled data to identify locations in a continuous space of the environment from the labeled data, wherein the dual regressor neural network comprises a feature extractor and a dual regressor, the feature extractor being configured to extract features from the labeled data, the dual regressor comprising two regressors having the same architecture, each regressor being trained to determine locations in the continuous space of the environment from features received from the feature extractor; During a second training phase, training the parameters of the dual regressor using the labeled data and the unlabeled data using the fixed parameters of the feature extractor trained during the first training phase, such that each of the two regressors recognizes the same labeled locations while processing the labeled data and recognizes different locations while processing the unlabeled data; and During a third training stage, parameters of the feature extractor are trained using an adversarial discriminator using the fixed parameters of the dual regressor trained using the labeled data and the unlabeled data during the second training stage to extract domain-invariant features from the unlabeled data having statistical properties of the labeled data according to the adversarial discriminator, so that each of the two regressors recognizes the same location while processing the domain-invariant features.

Images