Abstract: Systems and methods for knowledge distillation provide supervised training of a student network with a teacher network, including inputting a batch to the teacher network, inputting the batch to the student network, generating a teacher activation map at a layer of the teacher network, generating a student activation map at a layer of the student network corresponding to the layer of the teacher network, generating a pairwise teacher similarity matrix based on the teacher activation map, generating a pairwise student similarity matrix based on the student activation map, and minimizing a knowledge distillation loss defined as a difference between the pairwise teacher similarity matrix and the pairwise student similarity matrix.

Abstract: Systems and methods adapted for training a machine learning model to predict data labels are described. The approach includes receiving a first data set comprising first data objects and associated first data labels, and processing, with a user representation model, respective first data objects and associated data labels associated with a unique user representation by fusing the respective first data object and the associated first data labels. First data object representations of the respective first data objects are generated, and the first data object representations and the user representation model outputs are fused to create a user conditional object representation. The machine learning model updates corresponding parameters based on an error value based on a maximum similarity of the projections of the respective user conditional object representation and first data labels in a joint embedding space.

Abstract: A computer-implemented system and method for training a machine learning model are disclosed, the method includes: maintaining a data set representing a neural network having a plurality of weights; receiving input data comprising a plurality of time series data sets ending with timestamp t?1; generating, using the neural network and based on the input data, a probabilistic forecast distribution prediction at timestamp t and a selection value associated with the probabilistic forecast distribution prediction at timestamp t; computing a loss function based on the selection value; and updating at least one of the plurality of weights of the neural network based on the loss function.

Abstract: A computer implemented system for generating one or more data structures is described, the one or more data structures representing an unseen composition based on a first category and a second category observed individually in a training data set. During training of a generator, a proposed framework utilizes at least one of the following discriminators—three pixel-centric discriminators, namely, frame discriminator, gradient discriminator, video discriminator; and one object-centric relational discriminator. The three pixel-centric discriminators ensure spatial and temporal consistency across the frames, and the relational discriminator leverages spatio-temporal scene graphs to reason over the object layouts in videos ensuring the right interactions among objects.

Abstract: Systems and methods of generating interpretive data associated with data sets. Embodiments of systems may be for adapting Grad-CAM methods for embedding networks. The system includes a processor and a memory. The memory stores processor-executable instructions that, when executed, configure the processor to: obtain a subject data set; generate a feature embedding based on the subject data set; determine an embedding gradient weight based on a prior-trained embedding network and the feature embedding associated with the subject data set, the prior-trained embedding network defined based on a plurality of embedding gradient weights respectively corresponding to a feature map generated based on a plurality of training samples, and wherein the embedding gradient weight is determined based on querying a feature space for the feature embedding associated with the subject data set; and generate signals for communicating interpretive data associated with the embedding gradient weight.

Abstract: Variational Autoencoders (VAEs) have been shown to be effective in modeling complex data distributions. Conventional VAEs operate with fully-observed data during training. However, learning a VAE model from partially-observed data is still a problem. A modified VAE framework is proposed that can learn from partially-observed data conditioned on the fully-observed mask. A model described in various embodiments is capable of learning a proper proposal distribution based on the missing data. The framework is evaluated for both high-dimensional multimodal data and low dimensional tabular data.

Abstract: A computer implemented system for generating one or more data structures is described, the one or more data structures representing an unseen composition based on a first category and a second category observed individually in a training data set. During training of a generator, a proposed framework utilizes at least one of the following discriminators—three pixel-centric discriminators, namely, frame discriminator, gradient discriminator, video discriminator; and one object-centric relational discriminator. The three pixel-centric discriminators ensure spatial and temporal consistency across the frames, and the relational discriminator leverages spatio-temporal scene graphs to reason over the object layouts in videos ensuring the right interactions among objects.

Abstract: Systems and methods adapted for training a machine learning model to predict data labels are described. The approach includes receiving a first data set comprising first data objects and associated first data labels, and processing, with a user representation model, respective first data objects and associated data labels associated with a unique user representation by fusing the respective first data object and the associated first data labels. First data object representations of the respective first data objects are generated, and the first data object representations and the user representation model outputs are fused to create a user conditional object representation. The machine learning model updates corresponding parameters based on an error value based on a maximum similarity of the projections of the respective user conditional object representation and first data labels in a joint embedding space.

Abstract: Systems and methods for knowledge distillation provide supervised training of a student network with a teacher network, including inputting a batch to the teacher network, inputting the batch to the student network, generating a teacher activation map at a layer of the teacher network, generating a student activation map at a layer of the student network corresponding to the layer of the teacher network, generating a pairwise teacher similarity matrix based on the teacher activation map, generating a pairwise student similarity matrix based on the student activation map, and minimizing a knowledge distillation loss defined as a difference between the pairwise teacher similarity matrix and the pairwise student similarity matrix.

Abstract: Effectively training machine learning systems with incomplete/partial labels is a practical, technical problem that solutions described herein attempt to overcome. In particular, an approach to modify loss functions on a proportionality basis is noted in some embodiments. In other embodiments, a graph neural network is provided to help identify correlations/causations as between categories. In another set of embodiments, a prediction approach is described to, based on originally provided labels, predict labels for unlabelled training samples such that the proportion of labelled labels relative to all labels is increased.

Abstract: A variational auto-encoder model is trained to generate probabilities of action categories and probabilities of inter-arrival times of next action from a sequence of past actions by generating a concatenated representation of each action and associated time, encoding the concatenated representations, determining a conditional prior distribution for a next action, determining a conditional posterior distribution for the current action, sampling a latent variable from the conditional prior distribution, generating a probability distribution over a current action category, and generating a probability distribution over inter-arrival times for the current action category.