Abstract: The embodiments are directed towards providing personalized federated learning (PFL) models via sharable federated basis models. A model architecture and learning algorithm for PFL models is disclosed. The embodiments learn a set of basis models, which can be combined layer by layer to form a personalized model for each client using specifically learned combination coefficients. The set of basis models are shared with each client of a set of the clients. Thus, the set of basis models is common to each client of the set of clients. However, each client may generate a unique PFL based on their specifically learned combination coefficients. The unique combination of coefficients for each client may be encoded in a separate personalized vector for each of the clients.

Abstract: A computing system and method that can be used for safe and privacy preserving video representations of participants in a videoconference. In particular, the present disclosure provides a general pipeline for generating reconstructions of videoconference participants based on semantic statuses and/or activity statuses of the participants. The systems and methods of the present disclosure allow for videoconferences that convey necessary or meaningful information of participants through presentation of generalized representations of participants while filtering unnecessary or unwanted information from the representations by leveraging machine-learning models.

Abstract: The present disclosure provides systems, methods, and computer program products for modeling dependencies throughout a network using a global-self attention model with a content attention layer and a positional attention layer that operate in parallel. The model receives input data comprising content values and context positions. The content attention layer generates one or more output features for each context position based on a global attention operation applied to the content values independent of the context positions. The positional attention layer generates an attention map for each of the context positions based on one or more content values of the respective context position and associated neighboring positions. Output is determined based on the output features generated by the content attention layer and the attention map generated for each context position by the positional attention layer. The model improves efficiency and can be used throughout a deep network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing novel category discovery. One of the methods includes generating first local feature tensors from a first training image; obtaining previous local feature tensors generated from a previous training image; generating a first similarity tensor representing a similarity between the first local feature tensors and the previous local feature tensors; obtaining a second similarity tensor for a second training image; processing, using a neural network, the first training image to generate a first training output representing a class prediction for the first training image; obtaining a second training output representing a class prediction for the second training image; and generating an update to the neural network from (i) a similarity between the first similarity tensor and the second similarity tensor and (ii) a similarity between the first training output and the second training output.

Abstract: Systems and methods of the present disclosure are directed to a method for training a machine-learned visual attention model. The method can include obtaining image data that depicts a head of a person and an additional entity. The method can include processing the image data with an encoder portion of the visual attention model to obtain latent head and entity encodings. The method can include processing the latent encodings with the visual attention model to obtain a visual attention value and processing the latent encodings with a machine-learned visual location model to obtain a visual location estimation. The method can include training the models by evaluating a loss function that evaluates differences between the visual location estimation and a pseudo visual location label derived from the image data and between the visual attention value and a ground truth visual attention label.

Abstract: Systems and methods for training a restoration model can leverage training for two sub-tasks to train the restoration model to generate realistic and identity-preserved outputs. The systems and methods can balance the training of the generation task and the reconstruction task to ensure the generated outputs preserve the identity of the original subject while generating realistic outputs. The systems and methods can further leverage a feature quantization model and skip connections to improve the model output and overall training.

Abstract: Systems and methods for providing deep learning models capable of performing joint representation learning and new category discovery on a mixture of labeled and unlabeled data, which may include single- and multi-modal data. In some examples, a flexible end-to-end framework uses unified contrastive learning on labeled and unlabeled data based on both instance discrimination and category discrimination, and further uses Winner-Take-All hashing to generate a pseudo-label based on the similarity between each pair of unlabeled data points that can be used to train the model to generate clustering assignments for each unlabeled data point. In some examples, the unified contrastive learning may be further based on cross-modal discrimination.

Abstract: A computing system and method that can be used for safe and privacy preserving video representations of participants in a videoconference. In particular, the present disclosure provides a general pipeline for generating reconstructions of videoconference participants based on semantic statuses and/or activity statuses of the participants. The systems and methods of the present disclosure allow for videoconferences that convey necessary or meaningful information of participants through presentation of generalized representations of participants while filtering unnecessary or unwanted information from the representations by leveraging machine-learning models.