Lifelong learning or continual learning is the problem of training an AI agent continuously while also preventing it from forgetting its previously acquired knowledge. Streaming lifelong learning is a challenging setting of lifelong learning with the goal of continuous learning in a dynamic non-stationary environment without forgetting. We introduce a novel approach to lifelong learning, which is streaming (observes each training example only once), requires a single pass over the data, can learn in a class-incremental manner, and can be evaluated on-the-fly (anytime inference). To accomplish these, we propose a novel virtual gradients based approach for continual representation learning which adapts to each new example while also generalizing well on past data to prevent catastrophic forgetting. Our approach also leverages an exponential-moving-average-based semantic memory to further enhance performance. Experiments on diverse datasets with temporally correlated observations demonstrate our method’s efficacy and superior performance over existing methods.

In the task incremental learning problem, deep learning models suffer from catastrophic forgetting of previously seen classes/tasks as they are trained on new classes/tasks. This problem becomes even harder when some of the test classes do not belong to the training class set, i.e., the task incremental generalized zero-shot learning problem. We propose a novel approach to address the task incremental learning problem for both the non zero-shot and zero-shot settings. Our proposed approach, called Rectification-based Knowledge Retention (RKR), applies weight rectifications and affine transformations for adapting the model to any task. During testing, our approach can use the task label information (task-aware) to quickly adapt the network to that task. We also extend our approach to make it task-agnostic so that it can work even when the task label information is not available during testing. Specifically, given a continuum of test data, our approach predicts the task and quickly adapts the network to the predicted task. We experimentally show that our proposed approach achieves state-of-the-art results on several benchmark datasets for both non zero-shot and zero-shot task incremental learning.

Text-based emotion prediction is an important task in the field of affective computing. Most prior work has been restricted to predicting emotions corresponding to a few high-level emotion classes. This paper explores and experiments with various techniques for fine-grained (27 classes) emotion prediction†which appeared at ACII 2021. In particular, (1) we present a method to incorporate multiple annotations from different raters, (2) we analyze the model's performance on fused emotion classes and with sub-sampled training data, (3) we present a method to leverage the correlations among the emotion categories, and (4) we propose a new framework for text-based fine-grained emotion prediction through emotion definition modeling. The emotion definition-based model outperforms the existing state-of-the-art for fine-grained emotion dataset GoEmotions. The approach involves a multi-task learning framework that models definitions of emotions as an auxiliary task while being trained on the primary task of emotion prediction. We model definitions using masked language modeling and class definition prediction tasks. We show that this trained model can be used for transfer learning on other benchmark datasets in emotion prediction with varying emotion label sets, domains, and sizes. The proposed models outperform the baselines on transfer learning experiments demonstrating the model's generalization capability.

Test-time adaptation (TTA) is the problem of updating a pre-trained source model at inference time given test input(s) from a different target domain. Most existing TTA approaches assume the setting in which the target domain is stationary, i.e., all the test inputs come from a single target domain. However, in many practical settings, the test input distribution might exhibit a lifelong/continual shift over time. Moreover, existing TTA approaches also lack the ability to provide reliable uncertainty estimates, which is crucial when distribution shifts occur between the source and target domain. To address these issues, we present PETAL (Probabilistic lifElong Test-time Adaptation with self-training prior), which solves lifelong TTA using a probabilistic approach, and naturally results in (1) a student-teacher framework, where the teacher model is an exponential moving average of the student model, and (2) regularizing the model updates at inference time using the source model as a regularizer. To prevent model drift in the lifelong/continual TTA setting, we also propose a data-driven parameter restoration technique which contributes to reducing the error accumulation and maintaining the knowledge of recent domains by restoring only the irrelevant parameters. In terms of predictive error rate as well as uncertainty based metrics such as Brier score and negative log-likelihood, our method achieves better results than the current state-of-the-art for online lifelong test-time adaptation across various benchmarks, such as CIFAR-10C, CIFAR-100C, ImageNetC, and ImageNet3DCC datasets. The source code for our approach is accessible at https://github.com/dhanajitb/petal.

Allowing effective inference of latent vectors while training GANs can greatly increase their applicability in various downstream tasks. Recent approaches, such as ALI and BiGAN frameworks, develop methods of inference of latent variables in GANs by adversarially training an image generator along with an encoder to match two joint distributions of image and latent vector pairs. We generalize these approaches to incorporate multiple layers of feedback on reconstructions, self-supervision, and other forms of supervision based on prior or learned knowledge about the desired solutions. We achieve this by modifying the discriminator's objective to correctly identify more than two joint distributions of tuples of an arbitrary number of random variables consisting of images, latent vectors, and other variables generated through auxiliary tasks, such as reconstruction and inpainting or as outputs of suitable pre-trained models. We design a non-saturating maximization objective for the generator-encoder pair and prove that the resulting adversarial game corresponds to a global optimum that simultaneously matches all the distributions. Within our proposed framework, we introduce a novel set of techniques for providing self-supervised feedback to the model based on properties, such as patch-level correspondence and cycle consistency of reconstructions. Through comprehensive experiments, we demonstrate the efficacy, scalability, and flexibility of the proposed approach for a variety of tasks. The appendix of the paper can be found at the following link: https://drive.google.com/file/d/1i99e682CqYWMEDXlnqkqrctGLVA9viiz/view?usp=sharing

Many real-world classification problems often have classes with very few labeled training samples. Moreover, all possible classes may not be initially available for training, and may be given incrementally. Deep learning models need to deal with this two-fold problem in order to perform well in real-life situations. In this paper, we propose a novel Few-Shot Lifelong Learning (FSLL) method that enables deep learning models to perform lifelong/continual learning on few-shot data. Our method selects very few parameters from the model for training every new set of classes instead of training the full model. This helps in preventing overfitting. We choose the few parameters from the model in such a way that only the currently unimportant parameters get selected. By keeping the important parameters in the model intact, our approach minimizes catastrophic forgetting. Furthermore, we minimize the cosine similarity between the new and the old class prototypes in order to maximize their separation, thereby improving the classification performance. We also show that integrating our method with self-supervision improves the model performance significantly. We experimentally show that our method significantly outperforms existing methods on the miniImageNet, CIFAR-100, and CUB-200 datasets. Specifically, we outperform the state-of-the-art method by an absolute margin of 19.27% for the CUB dataset.

In this paper, we propose a new framework for fine-grained emotion prediction in the text through emotion definition modeling. Our approach involves a multi-task learning framework that models definitions of emotions as an auxiliary task while being trained on the primary task of emotion prediction. We model definitions using masked language modeling and class definition prediction tasks. Our models outperform existing state-of-the-art for fine-grained emotion dataset GoEmotions. We further show that this trained model can be used for transfer learning on other benchmark datasets in emotion prediction with varying emotion label sets, domains, and sizes. The proposed models outperform the baselines on transfer learning experiments demonstrating the generalization capability of the models.

Deep learning models suffer from catastrophic forgetting when trained in an incremental learning setting. In this work, we propose a novel approach to address the task incremental learning problem, which involves training a model on new tasks that arrive in an incremental manner. The task incremental learning problem becomes even more challenging when the test set contains classes that are not part of the train set, i.e., a task incremental generalized zero-shot learning problem. Our approach can be used in both the zero-shot and non zero-shot task incremental learning settings. Our proposed method uses weight rectifications and affine transformations in order to adapt the model to different tasks that arrive sequentially. Specifically, we adapt the network weights to work for new tasks by "rectifying" the weights learned from the previous task. We learn these weight rectifications using very few parameters. We additionally learn affine transformations on the outputs generated by the network in order to better adapt them for the new task. We perform experiments on several datasets in both zero-shot and non zero-shot task incremental learning settings and empirically show that our approach achieves state-of-the-art results. Specifically, our approach outperforms the state-of-the-art non zero-shot task incremental learning method by over 5% on the CIFAR-100 dataset. Our approach also significantly outperforms the state-of-the-art task incremental generalized zero-shot learning method by absolute margins of 6.91% and 6.33% for the AWA1 and CUB datasets, respectively. We validate our approach using various ablation studies.

As neural networks are increasingly being applied to real-world applications, mechanisms to address distributional shift and sequential task learning without forgetting are critical. Methods incorporating network expansion have shown promise by naturally adding model capacity for learning new tasks while simultaneously avoiding catastrophic forgetting. However, the growth in the number of additional parameters of many of these types of methods can be computationally expensive at larger scales, at times prohibitively so. Instead, we propose a simple task-specific feature map transformation strategy for continual learning, which we call Efficient Feature Transformations (EFTs). These EFTs provide powerful flexibility for learning new tasks, achieved with minimal parameters added to the base architecture. We further propose a feature distance maximization strategy, which significantly improves task prediction in class incremental settings, without needing expensive generative models. We demonstrate the efficacy and efficiency of our method with an extensive set of experiments in discriminative (CIFAR-100 and ImageNet-1K) and generative (LSUN, CUB-200, Cats) sequences of tasks. Even with low single-digit parameter growth rates, EFTs can outperform many other continual learning methods in a wide range of settings.

Continual Learning is a learning paradigm where learning systems are trained on a sequence of tasks. The goal here is to perform well on the current task without suffering from a performance drop on the previous tasks. Two notable directions among the recent advances in continual learning with neural networks are (1) variational Bayes based regularization by learning priors from previous tasks, and, (2) learning the structure of deep networks to adapt to new tasks. So far, these two approaches have been largely orthogonal. We present a novel Bayesian framework based on continually learning the structure of deep neural networks, to unify these distinct yet complementary approaches. The proposed framework learns the deep structure for each task by learning which weights to be used, and supports inter-task transfer through the overlapping of different sparse subsets of weights learned by different tasks. An appealing aspect of our proposed continual learning framework is that it is applicable to both discriminative (supervised) and generative (unsupervised) settings. Experimental results on supervised and unsupervised benchmarks demonstrate that our approach performs comparably or better than recent advances in continual learning.

We propose a novel approach for class incremental online learning in a limited data setting. This problem setting is challenging because of the following constraints: (1) Classes are given incrementally, which necessitates a class incremental learning approach; (2) Data for each class is given in an online fashion, i.e., each training example is seen only once during training; (3) Each class has very few training examples; and (4) We do not use or assume access to any replay/memory to store data from previous classes. Therefore, in this setting, we have to handle twofold problems of catastrophic forgetting and overfitting. In our approach, we learn robust representations that are generalizable across tasks without suffering from the problems of catastrophic forgetting and overfitting to accommodate future classes with limited samples. Our proposed method leverages the meta-learning framework with knowledge consolidation. The meta-learning framework helps the model for rapid learning when samples appear in an online fashion. Simultaneously, knowledge consolidation helps to learn a robust representation against forgetting under online updates to facilitate future learning. Our approach significantly outperforms other methods on several benchmarks.

We present a continual learning approach for generative adversarial networks (GANs), by designing and leveraging parameter-efficient feature map transformations. Our approach is based on learning a set of global and task-specific parameters. The global parameters are fixed across tasks whereas the task-specific parameters act as local adapters for each task, and help in efficiently obtaining task-specific feature maps. Moreover, we propose an element-wise addition of residual bias in the transformed feature space, which further helps stabilize GAN training in such settings. Our approach also leverages task similarities based on the Fisher information matrix. Leveraging this knowledge from previous tasks significantly improves the model performance. In addition, the similarity measure also helps reduce the parameter growth in continual adaptation and helps to learn a compact model. In contrast to the recent approaches for continually-learned GANs, the proposed approach provides a memory-efficient way to perform effective continual data generation. Through extensive experiments on challenging and diverse datasets, we show that the feature-map-transformation approach outperforms state-of-the-art methods for continually-learned GANs, with substantially fewer parameters. The proposed method generates high-quality samples that can also improve the generative-replay-based continual learning for discriminative tasks.

We present a meta-learning based generative model for zero-shot learning (ZSL) towards a challenging setting when the number of training examples from each seen class is very few. This setup contrasts with the conventional ZSL approaches, where training typically assumes the availability of a sufficiently large number of training examples from each of the seen classes. The proposed approach leverages meta-learning to train a deep generative model that integrates variational autoencoder and generative adversarial networks. We propose a novel task distribution where meta-train and meta-validation classes are disjoint to simulate the ZSL behaviour in training. Once trained, the model can generate synthetic examples from seen and unseen classes. Synthesize samples can then be used to train the ZSL framework in a supervised manner. The meta-learner enables our model to generates high-fidelity samples using only a small number of training examples from seen classes. We conduct extensive experiments and ablation studies on four benchmark datasets of ZSL and observe that the proposed model outperforms state-of-the-art approaches by a significant margin when the number of examples per seen class is very small.

Simple weighted averaging of word vectors often yields effective representations for sentences which outperform sophisticated seq2seq neural models in many tasks. While it is desirable to use the same method to represent documents as well, unfortunately, the effectiveness is lost when representing long documents involving multiple sentences. One of the key reasons is that a longer document is likely to contain words from many different topics; hence, creating a single vector while ignoring all the topical structure is unlikely to yield an effective document representation. This problem is less acute in single sentences and other short text fragments where the presence of a single topic is most likely. To alleviate this problem, we present P-SIF, a partitioned word averaging model to represent long documents. P-SIF retains the simplicity of simple weighted word averaging while taking a document's topical structure into account. In particular, P-SIF learns topic-specific vectors from a document and finally concatenates them all to represent the overall document. We provide theoretical justifications on the correctness of P-SIF. Through a comprehensive set of experiments, we demonstrate P-SIF's effectiveness compared to simple weighted averaging and many other baselines.

We present an attention-based ranking framework for learning to order sentences given a paragraph. Our framework is built on a bidirectional sentence encoder and a self-attention based transformer network to obtain an input order invariant representation of paragraphs. Moreover, it allows seamless training using a variety of ranking based loss functions, such as pointwise, pairwise, and listwise ranking. We apply our framework on two tasks: Sentence Ordering and Order Discrimination. Our framework outperforms various state-of-the-art methods on these tasks on a variety of evaluation metrics. We also show that it achieves better results when using pairwise and listwise ranking losses, rather than the pointwise ranking loss, which suggests that incorporating relative positions of two or more sentences in the loss function contributes to better learning.

We present a probabilistic framework for community discovery and link prediction for graph-structured data, based on a novel, gamma ladder variational autoencoder (VAE) architecture. We model each node in the graph via a deep hierarchy of gamma-distributed embeddings, and define each link probability via a nonlinear function of the bottom-most layer's embeddings of its associated nodes. In addition to leveraging the representational power of multiple layers of stochastic variables via the ladder VAE architecture, our framework offers the following benefits: (1) Unlike existing ladder VAE architectures based on real-valued latent variables, the gamma-distributed latent variables naturally result in non-negativity and sparsity of the learned embeddings, and facilitate their direct interpretation as membership of nodes into (possibly multiple) communities/topics; (2) A novel recognition model for our gamma ladder VAE architecture allows fast inference of node embeddings; and (3) The framework also extends naturally to incorporate node side information (features and/or labels). Our framework is also fairly modular and can leverage a wide variety of graph neural networks as the VAE encoder. We report both quantitative and qualitative results on several benchmark datasets and compare our model with several state-of-the-art methods.

Learning to classify unseen class samples at test time is popularly referred to as zero-shot learning (ZSL). If test samples can be from training (seen) as well as unseen classes, it is a more challenging problem due to the existence of strong bias towards seen classes. This problem is generally known as generalized zero-shot learning (GZSL). Thanks to the recent advances in generative models such as VAEs and GANs, sample synthesis based approaches have gained considerable attention for solving this problem. These approaches are able to handle the problem of class bias by synthesizing unseen class samples. However, these ZSL/GZSL models suffer due to the following key limitations: (i) Their training stage learns a class-conditioned generator using only seen class data and the training stage does not explicitly learn to generate the unseen class samples; (ii) They do not learn a generic optimal parameter which can easily generalize for both seen and unseen class generation; and (iii) If we only have access to a very few samples per seen class, these models tend to perform poorly. In this paper, we propose a meta-learning based generative model that naturally handles these limitations. The proposed model is based on integrating model-agnostic meta learning with a Wasserstein GAN (WGAN) to handle (i) and (iii), and uses a novel task distribution to handle (ii). Our proposed model yields significant improvements on standard ZSL as well as more challenging GZSL setting. In ZSL setting, our model yields 4.5%, 6.0%, 9.8%, and 27.9% relative improvements over the current state-of-the-art on CUB, AWA1, AWA2, and aPY datasets, respectively.

Many applications, such as text modelling, high-throughput sequencing, and recommender systems, require analysing sparse, high-dimensional, and overdispersed discrete (count or binary) data. Recent deep probabilistic models based on variational autoencoders (VAE) have shown promising results on discrete data but may have inferior modelling performance due to the insufficient capability in modelling overdispersion and model misspecification. To address these issues, we develop a VAE-based framework using the negative binomial distribution as the data distribution. We also provide an analysis of its properties vis-à-vis other models. We conduct extensive experiments on three problems from discrete data analysis: text analysis/topic modelling, collaborative filtering, and multi-label learning. Our models outperform state-of-the-art approaches on these problems, while also capturing the phenomenon of overdispersion more effectively.

We present an approach for lifelong/continual learning of convolutional neural networks (CNN) that does not suffer from the problem of catastrophic forgetting when moving from one task to the other. We show that the activation maps generated by the CNN trained on the old task can be calibrated using very few calibration parameters, to become relevant to the new task. Based on this, we calibrate the activation maps produced by each network layer using spatial and channel-wise calibration modules and train only these calibration parameters for each new task in order to perform lifelong learning. Our calibration modules introduce significantly less computation and parameters as compared to the approaches that dynamically expand the network. Our approach is immune to catastrophic forgetting since we store the task-adaptive calibration parameters, which contain all the task-specific knowledge and is exclusive to each task. Further, our approach does not require storing data samples from the old tasks, which is done by many replay based methods. We perform extensive experiments on multiple benchmark datasets (SVHN, CIFAR, ImageNet, and MS-Celeb), all of which show substantial improvements over state-of-the-art methods (e.g., a 29% absolute increase in accuracy on CIFAR-100 with 10 classes at a time). On large-scale datasets, our approach yields 23.8% and 9.7% absolute increase in accuracy on ImageNet-100 and MS-Celeb-10K datasets, respectively, by employing very few (0.51% and 0.35% of model parameters) task-adaptive calibration parameters.

In this work, we propose a modeling technique for jointly training image and video generation models by simultaneously learning to map latent variables with a fixed prior onto real images and interpolate over images to generate videos. The proposed approach models the variations in representations using residual vectors encoding the change at each time step over a summary vector for the entire video. We utilize the technique to jointly train an image generation model with a fixed prior along with a video generation model lacking constraints such as disentanglement. The joint training enables the image generator to exploit temporal information while the video generation model learns to flexibly share information across frames. Moreover, experimental results verify our approach's compatibility with pre-training on videos or images and training on datasets containing a mixture of both. A comprehensive set of quantitative and qualitative evaluations reveal the improvements in sample quality and diversity over both video generation and image generation baselines. We further demonstrate the technique's capabilities of exploiting similarity in features across frames by applying it to a model based on decomposing the video into motion and content. The proposed model allows minor variations in content across frames while maintaining the temporal dependence through latent vectors encoding the pose or motion features.

We present a domain adaptation based generative framework for zero-shot learning. Our framework addresses the problem of domain shift between the seen and unseen class distributions in zero-shot learning and minimizes the shift by developing a generative model trained via adversarial domain adaptation. Our approach is based on end-to-end learning of the class distributions of seen classes and unseen classes. To enable the model to learn the class distributions of unseen classes, we parameterize these class distributions in terms of the class attribute information (which is available for both seen and unseen classes). This provides a very simple way to learn the class distribution of any unseen class, given only its class attribute information, and no labeled training data. Training this model with adversarial domain adaptation further provides robustness against the distribution mismatch between the data from seen and unseen classes. Our approach also provides a novel way for training neural net based classifiers to overcome the hubness problem in zero-shot learning. Through a comprehensive set of experiments, we show that our model yields superior accuracies as compared to various state-of-the-art zero shot learning models, on a variety of benchmark datasets. Code for the experiments is available at github.com/vkkhare/ZSL-ADA.

We present a filter correlation based model compression approach for deep convolutional neural networks. Our approach iteratively identifies pairs of filters with the largest pairwise correlations and drops one of the filters from each such pair. However, instead of discarding one of the filters from each such pair naïvely, the model is re-optimized to make the filters in these pairs maximally correlated, so that discarding one of the filters from the pair results in minimal information loss. Moreover, after discarding the filters in each round, we further finetune the model to recover from the potential small loss incurred by the compression. We evaluate our proposed approach using a comprehensive set of experiments and ablation studies. Our compression method yields state-of-the-art FLOPs compression rates on various benchmarks, such as LeNet-5, VGG-16, and ResNet-50,56, while still achieving excellent predictive performance for tasks such as object detection on benchmark datasets.

We present a filter pruning approach for deep model compression, using a multitask network. Our approach is based on learning a a pruner network to prune a pre-trained target network. The pruner is essentially a multitask deep neural network with binary outputs that help identify the filters from each layer of the original network that do not have any significant contribution to the model and can therefore be pruned. The pruner network has the same architecture as the original network except that it has a multitask/multi-output last layer containing binary-valued outputs (one per filter), which indicate which filters have to be pruned. The pruner's goal is to minimize the number of filters from the original network by assigning zero weights to the corresponding output feature-maps. In contrast to most of the existing methods, instead of relying on iterative pruning, our approach can prune the network (original network) in one go and, moreover, does not require specifying the degree of pruning for each layer (and can learn it instead). The compressed model produced by our approach is generic and does not need any special hardware/software support. Moreover, augmenting with other methods such as knowledge distillation, quantization, and connection pruning can increase the degree of compression for the proposed approach. We show the efficacy of our proposed approach for classification and object detection tasks.

We present a label embedding based approach to large-scale multi-label learning, drawing inspiration from ideas rooted in distributional semantics, specifically the Skip Gram Negative Sampling (SGNS) approach, widely used to learn word embeddings. Besides leading to a highly scalable model for multi-label learning, our approach highlights interesting connections between label embedding methods commonly used for multi-label learning and paragraph embedding methods commonly used for learning representations of text data. The framework easily extends to incorporating auxiliary information such as label-label correlations; this is crucial especially when many training instances are only partially annotated. To facilitate end-to-end learning, we develop a joint learning algorithm that can learn the embeddings as well as a regression model that predicts these embeddings for the new input to be annotated, via efficient gradient based methods. We demonstrate the effectiveness of our approach through an extensive set of experiments on a variety of benchmark datasets, and show that the proposed models perform favorably as compared to state-of-the-art methods for large-scale multi-label learning.

Word embeddings have been widely adopted across several NLP applications.Most existing word embedding methods utilize sequential context of a word to learn its embedding.While there have been some attempts at utilizing syntactic context of a word, such methods result in an explosion of the vocabulary size.In this paper, we overcome this problem by proposing SynGCN, a flexible Graph Convolution based method for learning word embeddings.SynGCN utilizes the dependency context of a word without increasing the vocabulary size.Word embeddings learned by SynGCN outperform existing methods on various intrinsic and extrinsic tasks and provide an advantage when used with ELMo.We also propose SemGCN, an effective framework for incorporating diverse semantic knowledge for further enhancing learned word representations.We make the source code of both models available to encourage reproducible research.

We present a probabilistic framework for multi-label learning based on a deep generative model for the binary label vector associated with each observation. Our generative model learns deep multi-layer latent embeddings of the binary label vector, which are conditioned on the input features of the observation. The model also has an interesting interpretation in terms of a deep topic model, with each label vector representing a bag-of-words document, with the input features being its meta-data. In addition to capturing the structural properties of the label space (e.g., a near-low-rank label matrix), the model also offers a clean, geometric interpretation. In particular, the nonlinear classification boundaries learned by the model can be seen as the union of multiple convex polytopes. Our model admits a simple and scalable inference via efficient Gibbs sampling or EM algorithm. We compare our model with state-of-the-art baselines for multi-label learning on benchmark data sets, and also report some interesting qualitative results.

We present a novel deep learning architecture in which the convolution operation leverages heterogeneous kernels. The proposed HetConv (Heterogeneous Kernel-Based Convolution) reduces the computation (FLOPs) and the number of parameters as compared to standard convolution operation while still maintaining representational efficiency. To show the effectiveness of our proposed convolution, we present extensive experimental results on the standard convolutional neural network (CNN) architectures such as VGG and ResNet. We find that after replacing the standard convolutional filters in these architectures with our proposed HetConv filters, we achieve 3X to 8X FLOPs based improvement in speed while still maintaining (and sometimes improving) the accuracy. We also compare our proposed convolutions with group/depth wise convolutions and show that it achieves more FLOPs reduction with significantly higher accuracy.

Stochastic blockmodels (SBM) and their variants, $e.g.$, mixed-membership and overlapping stochastic blockmodels, are latent variable based generative models for graphs. They have proven to be successful for various tasks, such as discovering the community structure and link prediction on graph-structured data. Recently, graph neural networks, $e.g.$, graph convolutional networks, have also emerged as a promising approach to learn powerful representations (embeddings) for the nodes in the graph, by exploiting graph properties such as locality and invariance. In this work, we unify these two directions by developing a sparse variational autoencoder for graphs, that retains the interpretability of SBMs, while also enjoying the excellent predictive performance of graph neural nets. Moreover, our framework is accompanied by a fast recognition model that enables fast inference of the node embeddings (which are of independent interest for inference in SBM and its variants). Although we develop this framework for a particular type of SBM, namely the overlapping stochastic blockmodel, the proposed framework can be adapted readily for other types of SBMs. Experimental results on several benchmarks demonstrate encouraging results on link prediction while learning an interpretable latent structure that can be used for community discovery.

While convolutional neural networks (CNN) have achieved impressive performance on various classification/recognition tasks, they typically consist of a massive number of parameters. This results in significant memory requirement as well as computational overheads. Consequently, there is a growing need for filter-level pruning approaches for compressing CNN based models that not only reduce the total number of parameters but reduce the overall computation as well. We present a new min-max framework for filter-level pruning of CNNs. Our framework, called Play and Prune (PP), jointly prunes and fine-tunes CNN model parameters, with an adaptive pruning rate, while maintaining the model's predictive performance. Our framework consists of two modules: (1) An adaptive filter pruning (AFP) module, which minimizes the number of filters in the model; and (2) A pruning rate controller (PRC) module, which maximizes the accuracy during pruning. Moreover, unlike most previous approaches, our approach allows directly specifying the desired error tolerance instead of pruning level. Our compressed models can be deployed at run-time, without requiring any special libraries or hardware. Our approach reduces the number of parameters of VGG-16 by an impressive factor of 17.5X, and number of FLOPS by 6.43X, with no loss of accuracy, significantly outperforming other state-of-the-art filter pruning methods.

Paraphrase generation is an important problem in NLP, especially in question answering, information retrieval, information extraction, conversation systems, to name a few. In this paper, we address the problem of generating paraphrases automatically. Our proposed method is based on a combination of deep generative models (VAE) with sequence-to-sequence models (LSTM) to generate paraphrases, given an input sentence. Traditional VAEs when combined with recurrent neural networks can generate free text but they are not suitable for paraphrase generation for a given sentence. We address this problem by conditioning the both, encoder and decoder sides of VAE, on the original sentence, so that it can generate the given sentence's paraphrases. Unlike most existing models, our model is simple, modular and can generate multiple paraphrases, for a given sentence. Quantitative evaluation of the proposed method on a benchmark paraphrase dataset demonstrates its efficacy, and its performance improvement over the state-of-the-art methods by a significant margin, whereas qualitative human evaluation indicate that the generated paraphrases are well-formed, grammatically correct, and are relevant to the input sentence. Furthermore, we evaluate our method on a newly released question paraphrase dataset, and establish a new baseline for future research.

We present a deep generative model for Zero-Shot Learning (ZSL). Unlike most existing methods for this problem, that represent each class as a point (via a semantic embedding), we represent each seen/unseen class using a class-specific latent-space distribution, conditioned on class attributes. We use these latent-space distributions as a prior for a supervised variational autoencoder (VAE), which also facilitates learning highly discriminative feature representations for the inputs. The entire framework is learned end-to-end using only the seen-class training data. At test time, the label for an unseen-class test input is the class that maximizes the VAE lower bound. We further extend the model to a (i) semi-supervised/transductive setting by leveraging unlabeled unseen-class data via an unsupervised learning module, and (ii) few-shot learning where we also have a small number of labeled inputs from the unseen classes. We compare our model with several state-of-the-art methods through a comprehensive set of experiments on a variety of benchmark data sets.

We present a probabilistic, fully Bayesian framework for multi-label learning. Our framework is based on the idea of learning a joint low-rank embedding of the label matrix and the label co-occurrence matrix. The proposed framework has the following appealing aspects: (1) It leverages the sparsity in the label matrix and the feature matrix, which results in very efficient inference, especially for sparse datasets, commonly encountered in multi-label learning problems, and (2) By effectively utilizing the label co-occurrence information, the model yields improved prediction accuracies, especially in the case where the amount of training data is low and/or the label matrix has a significant fraction of missing labels. Our framework enjoys full local conjugacy and admits a simple inference procedure via a scalable Gibbs sampler. We report experimental results on a number of benchmark datasets, on which it outperforms several state-of-the-art multi-label learning models.

We present a generative framework for generalized zero-shot learning where the training and test classes are not necessarily disjoint. Built upon a variational autoencoder based architecture, consisting of a probabilistic encoder and a probabilistic conditional decoder, our model can generate novel exemplars from seen/unseen classes, given their respective class attributes. These exemplars can subsequently be used to train any off-the-shelf classification model. One of the key aspects of our encoder-decoder architecture is a feedback-driven mechanism in which a discriminator (a multivariate regressor) learns to map the generated exemplars to the corresponding class attribute vectors, leading to an improved generator. Our model's ability to generate and leverage examples from unseen classes to train the classification model naturally helps to mitigate the bias towards predicting seen classes in generalized zero-shot learning settings. Through a comprehensive set of experiments, we show that our model outperforms several state-of-the-art methods, on several benchmark datasets, for both standard as well as generalized zero-shot learning.

The latent feature relational model (LFRM) for graphs represents each node as having binary memberships in one or more communities. The community memberships can be represented in form of a binary vector and LFRM defines the link probability between any pair of nodes as a bilinear function of their community membership vectors. Moreover, using nonparametric Bayesian prior - Indian Buffet Process - on the community membership matrix enables learning the number of communities automatically from the data. However, despite its modeling flexibility, strong link predictive performance, and nice interpretability of binary embeddings, inference in LFRM remains a challenge and is typically done via MCMC or variational methods. These methods can be slow and may take a long time to converge. In this work, we apply the small variance asymptotics idea to the non-parametric Bayesian LFRM, utilizing the connection between exponential families and Bregman divergence. This leads to an overlapping k-means like objective function for the nonparametric Bayesian LFRM, which can be optimized using generic or specialized solvers. We also propose an iterative greedy algorithm to optimize the objective function and compare our approach with other inference methods on several benchmark datasets. Our results demonstrate that our inference algorithm is competitive to methods such as MCMC while being much faster.

We present a generative framework for zero-shot action recognition where some of the possible action classes do not occur in the training data. Our approach is based on modeling each action class using a probability distribution whose parameters are functions of the attribute vector representing that action class. In particular, we assume that the distribution parameters for any action class in the visual space can be expressed as a linear combination of a set of basis vectors where the combination weights are given by the attributes of the action class. These basis vectors can be learned solely using labeled data from the known (i.e., previously seen) action classes, and can then be used to predict the parameters of the probability distributions of unseen action classes. We consider two settings: (1) Inductive setting, where we use only the labeled examples of the seen action classes to predict the unseen action class parameters; and (2) Transductive setting which further leverages unlabeled data from the unseen action classes. Our framework also naturally extends to few-shot action recognition where a few labelled examples from unseen classes are available. Our experiments on benchmark datasets (UCF101, HMDB51 and Olympic) show significant performance improvements as compared to various baselines, in both standard zero-shot (disjoint seen and unseen classes) and generalized zero-shot learning settings.

We present a probabilistic framework for overlapping community discovery and link prediction for relational data, given as a graph. The proposed framework has: (1) a deep architecture which enables us to infer multiple layers of latent features/communities for each node, providing superior link prediction performance on more complex networks and better interpretability of the latent features; and (2) a regression model which allows directly conditioning the node latent features on the side information available in form of node attributes. Our framework handles both (1) and (2) via a clean, unified model, which enjoys full local conjugacy via data augmentation, and facilitates efficient inference via closed form Gibbs sampling. Moreover, inference cost scales in the number of edges which is attractive for massive but sparse networks. Our framework is also easily extendable to model weighted networks with count-valued edges. We compare with various state-of-the-art methods and report results, both quantitative and qualitative, on several benchmark data sets.

We present a scalable, generative framework for multi-label learning with missing labels. Our framework consists of a latent factor model for the binary label matrix, which is coupled with an exposure model to account for label missingness (i.e., whether a zero in the label matrix is indeed a zero or denotes a missing observation). The underlying latent factor model also assumes that the low-dimensional embeddings of each label vector are directly conditioned on the respective feature vector of that example. Our generative framework admits a simple inference procedure, such that the parameter estimation reduces to a sequence of simple weighted least-square regression problems, each of which can be solved easily, efficiently, and in parallel. Moreover, inference can also be performed in an online fashion using mini-batches of training examples, which makes our framework scalable for large data sets, even when using moderate computational resources. We report both quantitative and qualitative results for our framework on several benchmark data sets, comparing it with a number of state-of-the-art methods.

We present a probabilistic framework for efficient non-negative matrix factorization of discrete (count/binary) data with side-information. The side-information is given as a multi-level structure, taxonomy, or ontology, with nodes at each level being categorical-valued observations. For example, when modeling documents with a two-level side-information (documents being at level-zero), level-one may represent (one or more) authors associated with each document and level-two may represent affiliations of each author. The model easily generalizes to more than two levels (or taxonomy/ontology of arbitrary depth). Our model can learn embeddings of entities present at each level in the data/side-information hierarchy (e.g., documents, authors, affiliations, in the previous example), with appropriate sharing of information across levels. The model also enjoys full local conjugacy, facilitating efficient Gibbs sampling for model inference. Inference cost scales in the number of non-zero entries in the data matrix, which is especially appealing for real-world massive but sparse matrices. We demonstrate the effectiveness of the model on several real-world data sets.

We present a scalable probabilistic framework for learning from multi-relational data given in form of entity-relation-entity triplets, with a potentially massive number of entities and relations (e.g., in multi-relational networks, knowledge bases, etc.). We define each triplet via a relation-specific bilinear function of the embeddings of entities associated with it (these embeddings correspond to “topics”). To handle massive number of relations and the data sparsity problem (very few observations per relation), we also extend this model to allow sharing of parameters across relations, which leads to a substantial reduction in the number of parameters to be learned. In addition to yielding excellent predictive performance (e.g., for knowledge base completion tasks), the interpretability of our topic-based embedding framework enables easy qualitative analyses. Computational cost of our models scales in the number of positive triplets, which makes it easy to scale to massive real-world multi-relational data sets, which are usually extremely sparse. We develop simple-to-implement batch as well as online Gibbs sampling algorithms and demonstrate the effectiveness of our models on tasks such as multi-relational link-prediction, and learning from large knowledge bases.

Code variants represent alternative implementations of a computation, and are common in high-performance libraries and applications to facilitate selecting the most appropriate implementation for a specific execution context (target architecture and input dataset). Automating code variant selection typically relies on machine learning to construct a model during an offline learning phase that can be quickly queried at runtime once the execution context is known. In this paper, we define a new approach called architecture-adaptive code variant tuning, where the variant selection model is learned on a set of source architectures, and then used to predict variants on a new target architecture without having to repeat the training process. We pose this as a multi-task learning problem, where each source architecture corresponds to a task; we use device features in the construction of the variant selection model. This work explores the effectiveness of multi-task learning and the impact of different strategies for device feature selection. We evaluate our approach on a set of benchmarks and a collection of six NVIDIA GPU architectures from three distinct generations. We achieve performance results that are mostly comparable to the previous approach of tuning for a single GPU architecture without having to repeat the learning phase.

We present a probabilistic framework for learning with heterogeneous multiview data where some views are given as ordinal, binary, or real-valued feature matrices, and some views as similarity matrices. Our framework has the following distinguishing aspects: (i) a unified latent factor model for integrating information from diverse feature (ordinal, binary, real) and similarity based views, and predicting the missing data in each view, leveraging view correlations; (ii) seamless adaptation to binary/multiclass classification where data consists of multiple feature and/or similarity-based views; and (iii) an efficient, variational inference algorithm which is especially flexible in modeling the views with ordinal-valued data (by learning the cutpoints for the ordinal data), and extends naturally to streaming data settings. Our framework subsumes methods such as multiview learning and multiple kernel learning as special cases. We demonstrate the effectiveness of our framework on several real-world and benchmarks datasets.

We present a probabilistic model for tensor decomposition where one or more tensor modes may have side-information about the mode entities in form of their features and/or their adjacency network. We consider a Bayesian approach based on the Canonical PARAFAC (CP) decomposition and enrich this single-layer decomposition approach with a two-layer decomposition. The second layer fits a factor model for each layer-one factor matrix and models the factor matrix via the mode entities' features and/or the network between the mode entities. The second-layer decomposition of each factor matrix also learns a binary latent representation for the entities of that mode, which can be useful in its own right. Our model can handle both continuous as well as binary tensor observations. Another appealing aspect of our model is the simplicity of the model inference, with easy-to-sample Gibbs updates. We demonstrate the results of our model on several benchmarks datasets, consisting of both real and binary tensors.

We present a probabilistic framework for learning pairwise similarities between objects belonging to different modalities, such as drugs and proteins, or text and images. Our framework is based on learning a binary code based representation for objects in each modality, and has the following key properties: (i) it can leverage both pairwise as well as easy-to-obtain relative preference based cross-modal constraints, (ii) the probabilistic framework naturally allows querying for the most useful/informative constraints, facilitating an active learning setting (existing methods for cross-modal similarity learning do not have such a mechanism), and (iii) the binary code length is learned from the data. We demonstrate the effectiveness of the proposed approach on two problems that require computing pairwise similarities between cross-modal object pairs: cross-modal link prediction in bipartite graphs, and hashing based cross-modal similarity search.

We present a scalable Bayesian multi-label learning model based on learning low-dimensional label embeddings. Our model assumes that each label vector is generated as a weighted combination of a set of topics (each topic being a distribution over labels), where the combination weights (i.e., the embeddings) for each label vector are conditioned on the observed feature vector. This construction, coupled with a Bernoulli-Poisson link function for each label of the binary label vector, leads to a model with a computational cost that scales in the number of positive labels in the label matrix. This makes the model particularly appealing for real-world multi-label learning problems where the label matrix is usually very massive but highly sparse. Using a data-augmentation strategy leads to full local conjugacy in our model, facilitating simple and very efficient Gibbs sampling, as well as an Expectation Maximization algorithm for inference. Also, predicting the label vector at test time does not require doing an inference for the label embeddings and can be done in closed form. We report results on several benchmark data sets, comparing our model with various state-of-the art methods.

We present a scalable Bayesian framework for low-rank decomposition of multiway tensor data with missing observations. The key issue of pre-specifying the rank of the decomposition is sidestepped in a principled manner using a multiplicative gamma process prior. Both continuous and binary data can be analyzed under the framework, in a coherent way using fully conjugate Bayesian analysis. In particular, the analysis in the non-conjugate binary case is facilitated via the use of the Pólya-Gamma sampling strategy which elicits closed-form Gibbs sampling updates. The resulting samplers are efficient and enable us to apply our framework to large-scale problems, with time-complexity that is linear in the number of observed entries in the tensor. This is especially attractive in analyzing very large but sparsely observed tensors with very few known entries. Moreover, our method admits easy extension to the supervised setting where entities in one or more tensor modes have labels. Our method outperforms several state-of-the-art tensor decomposition methods on various synthetic and benchmark real-world datasets.

Stochastic block models provide a rich, probabilistic framework for modeling relational data by expressing the objects being modeled in terms of a latent vector representation. This representation can be a latent indicator vector denoting the cluster membership (hard clustering), a vector of cluster membership probabilities (soft clustering), or more generally a real-valued vector (latent space representation). Recently, a new class of overlapping stochastic block models has been proposed where the idea is to allow the objects to have hard memberships in multiple clusters (in form of a latent binary vector). This aspect captures the properties of many real-world networks in domains such as biology and social networks where objects can simultaneously have memberships in multiple clusters owing to the multiple roles they may have. In this paper, we improve upon this model in three key ways: (1) we extend the overlapping stochastic block model to the bipartite graph case which enables us to simultaneously learn the overlapping clustering of two different sets of objects in the graph, the unipartite graph is just a special case of our model, (2) we allow objects (in either set) to not have membership in any cluster by using a relevant object selection mechanism, and (3) we make use of additionally available object features (or a kernel matrix of pair wise object similarities) to further improve the overlapping clustering performance. We do this by explicitly encouraging similar objects to have similar cluster membership vectors. Moreover, using nonparametric Bayesian prior distributions on the key model parameters, we side-step the model selection issues such as selecting the number of clusters a priori. Our model is quite general and can be applied for both overlapping clustering and link prediction tasks in unipartite and bipartite networks (directed/undirected), or for overlapping co-clustering of general binary-valued data. Experiments on synthetic and real-world datasets from biology and social networks demonstrate that our model outperforms several state-of-the-art methods.

Multiple-output regression models require estimating multiple functions, one for each output. To improve parameter estimation in such models, methods based on structural regularization of the model parameters are usually needed. In this paper, we present a multiple-output regression model that leverages the covariance structure of the functions (i.e., how the multiple functions are related with each other) as well as the conditional covariance structure of the outputs. This is in contrast with existing methods that usually take into account only one of these structures. More importantly, unlike most of the other existing methods, none of these structures need be known a priori in our model, and are learned from the data. Several previously proposed structural regularization based multiple-output regression models turn out to be special cases of our model. Moreover, in addition to being a rich model for multiple-output regression, our model can also be used in estimating the graphical model structure of a set of variables (multivariate outputs) conditioned on another set of variables (inputs). Experimental results on both synthetic and real datasets demonstrate the effectiveness of our method.

Automatic clustering of Web pages helps a number of information retrieval tasks, such as improving user interfaces, collection clustering, introducing diversity in search results, etc. Typically, Web page clustering algorithms use only features extracted from the page-text. However, the advent of social-bookmarking Web sites, such as StumbleUpon.com and Delicious.com, has led to a huge amount of user-generated content such as the social tag information that is associated with the Web pages. In this article, we present a subspace based feature extraction approach that leverages the social tag information to complement the page-contents of a Web page for extracting beter features, with the goal of improved clustering performance. In our approach, we consider page-text and tags as two separate views of the data, and learn a shared subspace that maximizes the correlation between the two views. Any clustering algorithm can then be applied in this subspace. We then present an extension that allows our approach to be applicable even if the Web page corpus is only partially tagged, that is, when the social tags are present for not all, but only for a small number of Web pages. We compare our subspace based approach with a number of baselines that use tag information in various other ways, and show that the subspace based approach leads to improved performance on the Web page clustering task. We also discuss some possible future work including an active learning extension that can help in choosing which Web pages to get tags for, if we only can get the social tags for only a small number of Web pages.

Perimeter distinction in a wireless network is the ability to distinguish locations belonging to different perimeters. It is complementary to existing localization techniques. A draw-back of the localization method is that when a transmitter is at the edge of an area, an algorithm with isotropic error will estimate its location in the wrong area at least half of the time. In contrast, perimeter distinction classifies the location as being in one area or the adjacent regardless of the transmitter position within the area. In this paper, we use the naturally different wireless fading conditions to accurately distinguish locations across perimeters. We examine the use of two types of wireless measurements: received signal strength (RSS) and wireless link signature (WLS), and propose multiple methods to retain good distinction rates even when the receiver faces power manipulation by malicious transmitters. Using extensive measurements of indoor and outdoor perimeters, we find that WLS outperforms RSS in various fading conditions. Even without using signal power WLS can achieve accurate perimeter distinction up to 80%. When we train our perimeter distinction method with multiple measurements within the same perimeter, we show that we are able to improve the accuracy of perimeter distinction, up to 98%.

We consider a general inference setting for discrete probabilistic graphical models where we seek maximum a posteriori (MAP) estimates for a subset of the random variables (max nodes), marginalizing over the rest (sum nodes). We present a hybrid message-passing algorithm to accomplish this. The hybrid algorithm passes a mix of sum and max messages depending on the type of source node (sum or max). We derive our algorithm by showing that it falls out as the solution of a particular relaxation of a variational framework. We further show that the Expectation Maximization algorithm can be seen as an approximation to our algorithm. Experimental results on synthetic and real-world datasets, against several baselines, demonstrate the efficacy of our proposed algorithm.

In many clustering problems, we have access to multiple views of the data each of which could be individually used for clustering. Exploiting information from multiple views, one can hope to find a clustering that is more accurate than the ones obtained using the individual views. Since the true clustering would assign a point to the same cluster irrespective of the view, we can approach this problem by looking for clusterings that are consistent across the views, i.e., corresponding data points in each view should have same cluster membership. We propose a spectral clustering framework that achieves this goal by co-regularizing the clustering hypotheses, and propose two co-regularization schemes to accomplish this. Experimental comparisons with a number of baselines on two synthetic and three real-world datasets establish the efficacy of our proposed approaches.

Canonical Correlation Analysis (CCA) is a useful technique for modeling dependencies between two (or more) sets of variables. Building upon the recently suggested probabilistic interpretation of CCA, we propose a nonparametric, fully Bayesian framework that can automatically select the number of correlation components, and effectively capture the sparsity underlying the projections. In addition, given (partially) labeled data, our algorithm can also be used as a (semi)supervised dimensionality reduction technique, and can be applied to learn useful predictive features in the context of learning a set of related tasks. Experimental results demonstrate the efficacy of the proposed approach for both CCA as a stand-alone problem, and when applied to multi-label prediction.

We propose a nonparametric Bayesian factor regression model that accounts for uncertainty in the number of factors, and the relationship between factors. To accomplish this, we propose a sparse variant of the Indian Buffet Process and couple this with a hierarchical model over factors, based on Kingman's coalescent. We apply this model to two problems (factor analysis and factor regression) in gene-expression data analysis.