543 results about "Hyperparameter" patented technology

A variational Relevance Vector Machine (RVM) is disclosed. The RVM is a probabilistic basis model. Sparsity is achieved through a Bayesian treatment, where a prior is introduced over the weights governed by a set of what are referred to as hyperparameters—one such hyperparameter associated with each weight. An approximation to the joint posterior distribution over weights and hyperparameters is iteratively estimated from the data. The posterior distribution of many of the weights is sharply peaked around zero, in practice. The variational RVM utilizes a variational approach to solve the model, in particular using product approximations to obtain the posterior distribution.

The invention discloses a particle filtering method based on a Gaussian mixture model and variational Bayes, comprising the following steps: (1) modeling observation noise using a Gaussian mixture model, and initializing the initial state; (2) randomly generating N initial particles based on the probability density function of the initial state; (3) initializing the super parameters of the unknown parameters in the Gaussian mixture model of observation noise; (4) generating sampling particles from a chosen importance reference function; (5) updating measurement, and calculating the particle weights according to the latest observation value and a particle weight iteration formula; (7) using a variational Bayesian method to get the distribution of the unknown parameters in the Gaussian mixture model by means of loop iteration; and (7) normalizing the particle weights, and re-sampling a particle set in view of particle degradation. Through the scheme, the filtering precision and the target state estimation performance are improved effectively.

Embodiments of the present invention provide methods and systems for performing automatic diarisation of sound recordings including speech from one more speakers. The automatic diarisation has a development or training phase and a utilisation or evaluation phase. In the development or training phase background models and hyperparameters are generated from already annotated sound recordings. These models and hyperparameters are applied during the evaluation or utilisation phase to diarise new or not previously diarised or annotated recordings.

A method for modelling a dataset includes a training phase, wherein the dataset is applied to a non-stationary Gaussian process kernel in order to optimize the values of a set of hyperparameters associated with the Gaussian process kernel, and an evaluation phase in which the dataset and Gaussian process kernel with optimized hyperparameters are used to generate model data. The evaluation phase includes a nearest neighbour selection step. The method may include generating a model at a selected resolution.

A method of computerised data analysis and synthesis is described. First and second datasets of a quantity of interest are stored. A Gaussian process model is generated using the first and second datasets to compute optimized kernel and noise hyperparameters. The Gaussian process model is applied using the stored first and second datasets and hyperparameters to perform Gaussian process regression to compute estimates of unknown values of the quantity of interest. The resulting computed estimates of the quantity of interest result from a non-parametric Gaussian process fusion of the first and second measurement datasets. The first and second datasets may be derived from the same or different measurement sensors. Different sensors may have different noise and / or other characteristics.

The invention provides a hyper-parameter automatic optimization method and system of non-supervised machine learning. The hyper-parameter automatic optimization method of non-supervised machine learning includes the steps: according to the non-supervised machine learning algorithm, determining an algorithm performance assessment model of the non-supervised machine learning algorithm, the hyper-parameter of the non-supervised machine learning algorithm, the searching space of the hyper-parameter and the evaluation criterion of hyper-parameter optimization; and according to the algorithm performance assessment model, the searching space and the evaluation criterion, determining the optimal value of the hyper-parameter. The hyper-parameter automatic optimization method and system of non-supervised machine learning can deeply analyze the hyper-parameter problem in the non-supervised machine learning algorithm, can analyze the distribution rules of hyper-parameter in the algorithm, can assess the learning effect under different hyper-parameters, and can apply the rules to model training of machine learning so as to achieve the aim of automatically selecting the suitable hyper-parameter,and then the whole optimization process of hyper-parameter is automatically completed, so that the optimization efficiency is high and the usage complexity of the algorithm can be greatly reduced.

The invention discloses a long-term and short-term memory neural network based flight delay grading early warning method. The method includes analyzing an aviation meteorological message so that aviation meteorological data required by flight delay prediction can be obtained; performing multi-source data fusion to form an initial flight delay data set; converting non-numerical data into numericaldata by using semantic transformation, performing grading prediction on delay characteristics, and performing discrete partition on type characteristics and weather characteristics; performing data cleaning, missing value complementing and normalized processing to form a flight delay grading prediction standard data set, and performing partition; training long-term and short-term memory neural network based flight delay grading prediction models in batches on a training set; obtaining a long-term and short-term memory neural network model having the optimal hyperparameter on a verification set; performing verification on the performance of the optimal flight delay grading prediction model on a test set; and determining a delay early warning grade according to obtained flight delay grades through prediction. The method can effectively enhance the accuracy and reliability of flight delay early warning.

The invention provides a small area-level ultra-short-term load prediction and visualization method based on a Long Short-Term Memory (LSTM) network, the method comprising: step 1: an input and outputvariable of the model is determined; step 2: The input and output data sets are separately preprocessed; step 3: the depth LSTM load prediction model is constructed and the random search method is used to find the appropriate hyperparameters until the test set prediction error is minimized. step 4: The t-SNE visualization technology is used to visually characterize the network hidden layer vector, and the correlation coefficient heat map is formed according to the hidden layer vector to perform correlation quantitative analysis, thereby reflecting the network's ability to extract feature datafrom the input data. The method aims to utilize the feature extraction capabilities of the deep learning model and the LSTM temporal correlation learning capabilities to achieve higher prediction accuracy than the machine learning model.

The disclosed embodiments provide a system for performing online hyperparameter tuning in distributed machine learning. During operation, the system uses input data for a first set of versions of a statistical model for a set of entities to calculate a batch of performance metrics for the first set of versions. Next, the system applies an optimization technique to the batch to produce updates to a set of hyperparameters for the statistical model. The system then uses the updates to modulate the execution of a second set of versions of the statistical model for the set of entities. When a new entity is added to the set of entities, the system updates the set of hyperparameters with a new dimension for the new entity.

The embodiment of the invention provides an interactive automated big data analysis system and method. The interactive automated big data analysis system includes a data receiving module, an operatormanagement module, an analysis model generation module, a process analysis and optimization module and a task scheduling execution module. First, the system supports an interactive function with strong ease of use; second, the system provides users with functions such as batch build process, differentiated multiplexing process, and dynamic planning process through flow control operators, which greatly simplifies the building steps of complex analysis processes. In particular, the process analysis and optimization module is used to perform algorithm selection for each analysis process, users donot have to consider the algorithm implementation problem, and the users' use threshold in the data analysis technical aspect is greatly reduced; in addition, the process analysis and optimization module is used to perform hyperparameter selection for processes needing to undergo hyperparameter optimization. The system automatically selects optimal execution hyperparameters for tasks with uncertain parameters, making the construction and execution of the analysis process more concise and efficient.

The invention provides a method, a device, an apparatus and a medium for optimizing super parameters in a machine learning model. The method comprises: acquiring a template code for training a machinelearning model, wherein the template code identifies a respective set of values of each of one or more hyperparameters; parsing the template code to identify the respective set of values of one or more super parameters; generating at least partial combinations of values of one or more super parameters based on respective sets of values of the one or more super parameters; acquiring an evaluationresult on a model effect obtained by training a machine learning model according to each of at least partial value combinations; and determining an optimal value of one or more super-parameters of themachine learning model based on the evaluation results. Thus, the consumption of manual parameter adjustment can be reduced, and the development efficiency of machine learning (such as deep learning)can be effectively improved.

A method for training a disturbance rejection model that is configured to model an operation of a system so to calculate a predicted value for a system output at a future time. The disturbance rejection model may include a network for mapping system inputs to the system output, the network including a weight vector and a feedback coefficient. The method may include: obtaining a training dataset, and training the disturbance rejection model pursuant to the training dataset. The training may include calculating updated values for each of the weight vector and the feedback coefficient of the network by minimizing an error function that include a first hyperparameter and a second hyperparameter. The first hyperparameter may include a vector for penalizing the weight vector and the second hyperparameter may include a scalar.

The present invention provides a method for integrating crowdsourced annotations. According to the method provided by the present invention, a generalized anti-Gaussian distribution is defined by using a regularization hyperparameter, a spacing distance hyperparameter, an annotator voting weight, and a difference between the number of times that annotators annotate a current prediction item as a corresponding estimate and the number of times that annotators annotate the current prediction item as a subcategory; sampling is performed to obtain an auxiliary parameter; and annotator weights are updated by using the auxiliary parameter, thereby significantly enhancing a discrimination capability of a model. Then, a traditional annotation integration majority voting model and a confusion matrix model are integrated, thereby achieving an objective of more comprehensively describing a data generation process. In addition, an accurate prediction item update value is obtained through sampling, and moreover, running efficiency is also improved.

The invention relates to the field of cognitive radio communication and provides a multi-layer distributed combined spectrum sensing algorithm based on a Dirichlet process so as to realize dynamic spectrum sensing. The sensing data acquired by secondary users in a plurality of hierarchical centers is fused to search the optimized sensing information. The automatic data packet is realized by employing the Dirichlet process, a shared hyper-parameter and a corresponding divergent probability in each packet are estimated by a bayesian model, the hyper-parameter is acquired by employing a standard Viterbi algorithm, and the hyper-parameter is compared with a decision threshold value to acquire a final spectrum decision result so as to determine whether a channel is available. Due to the design, the space diversity information of the compressed sensing data is fully considered, and the uncertainty of an individual secondary user on the compressed sensing data is reduced, so that the normalized mean squared error performance is high, the compressed sensing data information in each hierarchical center can be effectively obtained through the algorithm, high accurate detection probability and low false alarm probability are acquired, and the spectrum sensing performance is improved.

Existing methods for analyzing person-specific ‘connectomes’ are not computationally equipped for scalable, flexible, and integrated processing across multiple network resolutions and drawing from disparate data modalities—a major roadblock to utilizing ensembles and hierarchies of connectomes to solve person-specific machine-learning problems. The processes implemented in software described herein consists of an end-to-end pipeline for deploying ensembles and hierarchies of network-generating workflows that can utilize multimodal, person-specific data to sample networks, extracted from that data, across a grid of network-defining hyperparameters. In essence, this pipeline enables users to perform ensemble sampling of connectomes for given individual(s) based on any input phenotypic datatype, constructed from any data modality or hierarchy of modalities at any scale, and based on any set of network-defining hyperparameters.

The invention discloses a support vector regression machine model selection method, and brings forward a novel support vector regression machine model selection method based on a hybrid kernel function and volume Kalman filtering, for solving the problem of model selection of a support vector regression machine. The hybrid kernel function is selected as a kernel function of the support vector machine, a combination coefficient of the hybrid kernel function is embedded into a superparametric state vector composed of a kernel function parameter and a regression parameter, accordingly, the problem of model selection is converted into a problem of state estimation of a nonlinear system, and then based on the high-performance volume Kalman filtering, superparametric estimation is carried out. A simulation experiment shows that the method brought forward by the invention, compared to a volume Kalman filtering support vector regression machine model selection method of a single kernel function and a genetic algorithm, has the following advantages: the generalization capability of a decision regression function obtained through the method is greater, and the prediction precision is higher.

Systems and methods for implementing an application programming interface (API) that controls operations of a machine learning tuning service for tuning a machine learning model for improved accuracy and computational performance includes an API that is in control communication the tuning service that: executes a first API call function that includes an optimization work request that sets tuning parameters for tuning hyperparameters of a machine learning model; and initializes an operation of distinct tuning worker instances of the service that each execute distinct tuning tasks for tuning the hyperparameters; executes a second API call function that identifies raw values for the hyperparameters; and generates suggestions comprising proposed hyperparameter values selected from the plurality of raw values for each of the hyperparameters; and executes a third API call function that returns performance metrics relating to a real-world performance of the subscriber machine learning model executed with the proposed hyperparameter values.

The invention relates to the technical field of atmosphere and ocean science, and discloses a sea fog level intelligent forecasting method and system, which comprises the following steps: collecting a numerical mode forecasting result and conventional observation data of a meteorological station, and fusing and controlling the quality of the collected data; the method comprises the following steps: extracting key meteorological elements influencing visibility as forecasting factors by using feature analysis methods such as Pearson correlation coefficient test, causal correlation test and time lag analysis, and meanwhile, using visibility observation and geographical time factors in a period of time before mode report start as auxiliary forecasting factors; correcting key meteorological elements in a numerical mode forecasting result by adopting a machine learning correction method based on a large amount of site observation data; a sea fog intelligent forecasting model is built and optimized by adopting technologies such as a machine learning algorithm, hyper-parameter automatic tuning and integrated learning; and carrying out grade forecasting on the visibility by using the sea fog intelligent forecasting model, realizing site forecasting and grid forecasting of the visibility, and checking the forecasting accuracy.

An apparatus for generating a temperature prediction model is disclosed. The apparatus for generating a temperature prediction model includes the temperature prediction model configured to provide a simulation environment, and a processor configured to set a hyperparameter of the temperature prediction model, train the temperature prediction model, in which the hyperparameter is set, so that the temperature prediction model, in which the hyperparameter is set, outputs a predicted temperature, update the hyperparameter on the basis of a difference between the predicted temperature, which is outputted from the trained temperature prediction model, and an actual temperature, and repeat the setting of the hyperparameter, the training of the temperature prediction model, and the updating of the hyperparameter on the basis of the difference between the predicted temperature and the actual temperature by a predetermined number of times or more to set a final hyperparameter of the temperature prediction model.

A simple yet powerful Bayesian model of linear regression is disclosed for methods and systems of machine learning. Unlike previous treatments that have either considered finding hyperparameters through maximum likelihood or have used a simple prior that makes the computation tractable but can lead to overfitting in high dimensions, the disclosed methods use a combination of linear algebra and numerical integration to work a full posterior over hyperparameters in a model with a prior that naturally avoids overfitting. The resulting algorithm is efficient enough to be practically useful. The approach can be viewed as a fully Bayesian version of the discriminative regularized least squares algorithm.

A system and method for lossy image and video compression and transmission that utilizes a neural network as a function to map a known noise image to a desired or target image, allowing the transfer only of hyperparameters of the function instead of a compressed version of the image itself. This allows the recreation of a high-quality approximation of the desired image by any system receiving the hyperparameters, provided that the receiving system possesses the same noise image and a similar neural network. The amount of data required to transfer an image of a given quality is dramatically reduced versus existing image compression technology. Being that video is simply a series of images, the application of this image compression system and method allows the transfer of video content at rates greater than existing technologies in relation to the same image quality.

Methods and systems are provided to determine suitable hyperparameters for a machine learning model and / or feature engineering process. A suitable machine learning model and associated hyperparameters are determined by analyzing a dataset. Suitable hyperparameter values for compatible machine learning models having one or more hyperparameters in common and a compatible dataset schema are identified. Hyperparameters may be ranked according to each of their respective influences on a model performance metrics, and hyperparameter values identified as having greater influence may be more aggressively searched.

The invention provides a GPR lithium battery health state prediction method based on a neural network kernel function. The method comprises the steps that a covariance function is determined based on the neural network kernel function to construct a GPR prediction model; a mean function in the GPR prediction model and hyperparameters in the covariance function are initialized; a logarithmic maximum likelihood estimation function is used to optimize the hyperparameters; and training data and test data are input into the GPR prediction model to acquire the value of the test data. According to the lithium battery health state prediction method provided by the invention, the accuracy and the precision of battery SOH value prediction are high, and the uncertainty is reduced.

A method for modelling a dataset includes a training phase, wherein the dataset is applied to a non-stationary Gaussian process kernel in order to optimize the values of a set of hyperparameters associated with the Gaussian process kernel, and an evaluation phase in which the dataset and Gaussian process kernel with optimized hyperparameters are used to generate model data. The evaluation phase includes a nearest neighbor selection step. The method may include generating a model at a selected resolution.

A method for auto-labeling a training image to be used for learning a neural network is provided for achieving high precision. The method includes steps of: an auto-labeling device (a) instructing a meta ROI detection network to generate a feature map and to acquire n current meta ROIs, on the specific training image, grouped according to each of locations of each of the objects; and (b) generating n manipulated images by cropping regions, corresponding to the n current meta ROIs, on the specific training image, instructing an object detection network to output each of n labeled manipulated images having each of bounding boxes for each of the n manipulated images, and generating a labeled specific training image by merging the n labeled manipulated images. The method can be performed by using an online learning, a continual learning, a hyperparameter learning, and a reinforcement learning with policy gradient algorithms.

The invention provides a first-visual-angle interactive action recognition method based on global and local network fusion. The method comprises: a video is sampled to obtain different actions to obtain images and thus an action sample is formed; dimension unification processing is carried out on an action segment obtained by sampling, data enhancement is carried out, a 3D convolutional network based on a global image as an input is trained, and global spatio-temporal features of the action are learned to obtain a network classification model; a local area with significance action occurrence in the action segment is located by using a sparse optical flow; after dimension unification processing of local areas of different actions, hyperparameters of the network are adjusted, a 3D convolutional network based on a global image as an input is trained, local significant action features are learned to obtain a network classification model; and action samples are obtained by multiple samplingon the same video, statistics and ranking of the prediction numbers provided by the global and local models are carried out based on a voting method, and the type with the largest prediction numbersis used as an identified action tag.

The invention relates to an electric energy metering device abnormality detecting method based on a long and short term memory model. The method comprises the steps of (1) acquiring original collecteddata, (2) carrying out data preprocessing, (3) setting model hyperparameters, (4) training the model with a training sample, (5) testing the model with a test sample, and (6) output an electric energy metering device abnormality detecting result. According to the electric energy metering device abnormality detecting method based on the long and short term memory model, the deep learning theory isapplied to digital electric energy meter abnormality detection, and the change characteristics of each piece of collected data under various faults are automatically learned in case of a large amountof data and easy missing, at the same time, the method has good fault tolerance, and the improvement of the accuracy and timeliness of the abnormality detection of an electric energy metering deviceis helped.

A method for segmentation of underground drainage pipeline defects based on full convolutional neural network includes steps of: collecting a data set of the underground drainage pipeline defects; processing the data set of the underground drainage pipeline defects; optimizing with a semantic segmentation algorithm; adjusting model hyperparameters; training a model; verifying the model; and testing the model. The method adopts a deep learning algorithm, optimizes the FCN full convolutional neural network, develops a semantic segmentation method suitable for complex and similar defect characteristics of underground drainage pipelines, and adopts real underground drainage pipeline defect detection big data, thereby realizing pixel-level segmentation of the underground drainage pipeline defects and providing better robustness and generality. The detection accuracy and efficiency of the underground drainage pipeline defects are effectively improved.