Gentle robustness implies Generalization

Robustness and generalization ability of machine learning models are of utmost importance in various application domains. There is a wide interest in efficient ways to analyze those properties. One important direction is to analyze connection between those two properties. Prior theories suggest that a robust learning algorithm can produce trained models with a high generalization ability. However, we show in this work that the existing error bounds are vacuous for the Bayes optimal classifier which is the best among all measurable classifiers for a classification problem with overlapping classes. Those bounds cannot converge to the true error of this ideal classifier. This is undesirable, surprizing, and never known before. We then present a class of novel bounds, which are model-dependent and provably tighter than the existing robustness-based ones. Unlike prior ones, our bounds are guaranteed to converge to the true error of the best classifier, as the number of samples increases. We further provide an extensive experiment and find that two of our bounds are often non-vacuous for a large class of deep neural networks, pretrained from ImageNet.

On Inference Stability for Diffusion Models

Denoising Probabilistic Models (DPMs) represent an emerging domain of generative models that excel in generating diverse and high-quality images. However, most current training methods for DPMs often neglect the correlation between timesteps, limiting the model's performance in generating images effectively. Notably, we theoretically point out that this issue can be caused by the cumulative estimation gap between the predicted and the actual trajectory. To minimize that gap, we propose a novel \textit{sequence-aware} loss that aims to reduce the estimation gap to enhance the sampling quality. Furthermore, we theoretically show that our proposed loss function is a tighter upper bound of the estimation loss in comparison with the conventional loss in DPMs. Experimental results on several benchmark datasets including CIFAR10, CelebA, and CelebA-HQ consistently show a remarkable improvement of our proposed method regarding the image generalization quality measured by FID and Inception Score compared to several DPM baselines. Our code and pre-trained checkpoints are available at \url{https://github.com/viettmab/SA-DPM}.

KOPPA: Improving Prompt-based Continual Learning with Key-Query Orthogonal Projection and Prototype-based One-Versus-All

Drawing inspiration from prompt tuning techniques applied to Large Language Models, recent methods based on pre-trained ViT networks have achieved remarkable results in the field of Continual Learning. Specifically, these approaches propose to maintain a set of prompts and allocate a subset of them to learn each task using a key-query matching strategy. However, they may encounter limitations when lacking control over the correlations between old task queries and keys of future tasks, the shift of features in the latent space, and the relative separation of latent vectors learned in independent tasks. In this work, we introduce a novel key-query learning strategy based on orthogonal projection, inspired by model-agnostic meta-learning, to enhance prompt matching efficiency and address the challenge of shifting features. Furthermore, we introduce a One-Versus-All (OVA) prototype-based component that enhances the classification head distinction. Experimental results on benchmark datasets demonstrate that our method empowers the model to achieve results surpassing those of current state-of-the-art approaches by a large margin of up to 20%.

Continual Learning with Optimal Transport based Mixture Model

Online Class Incremental learning (CIL) is a challenging setting in Continual Learning (CL), wherein data of new tasks arrive in incoming streams and online learning models need to handle incoming data streams without revisiting previous ones. Existing works used a single centroid adapted with incoming data streams to characterize a class. This approach possibly exposes limitations when the incoming data stream of a class is naturally multimodal. To address this issue, in this work, we first propose an online mixture model learning approach based on nice properties of the mature optimal transport theory (OT-MM). Specifically, the centroids and covariance matrices of the mixture model are adapted incrementally according to incoming data streams. The advantages are two-fold: (i) we can characterize more accurately complex data streams and (ii) by using centroids for each class produced by OT-MM, we can estimate the similarity of an unseen example to each class more reasonably when doing inference. Moreover, to combat the catastrophic forgetting in the CIL scenario, we further propose Dynamic Preservation. Particularly, after performing the dynamic preservation technique across data streams, the latent representations of the classes in the old and new tasks become more condensed themselves and more separate from each other. Together with a contraction feature extractor, this technique facilitates the model in mitigating the catastrophic forgetting. The experimental results on real-world datasets show that our proposed method can significantly outperform the current state-of-the-art baselines.

Face Swapping as A Simple Arithmetic Operation

We propose a novel high-fidelity face swapping method called "Arithmetic Face Swapping" (AFS) that explicitly disentangles the intermediate latent space W+ of a pretrained StyleGAN into the "identity" and "style" subspaces so that a latent code in W+ is the sum of an "identity" code and a "style" code in the corresponding subspaces. Via our disentanglement, face swapping (FS) can be regarded as a simple arithmetic operation in W+, i.e., the summation of a source "identity" code and a target "style" code. This makes AFS more intuitive and elegant than other FS methods. In addition, our method can generalize over the standard face swapping to support other interesting operations, e.g., combining the identity of one source with styles of multiple targets and vice versa. We implement our identity-style disentanglement by learning a neural network that maps a latent code to a "style" code. We provide a condition for this network which theoretically guarantees identity preservation of the source face even after a sequence of face swapping operations. Extensive experiments demonstrate the advantage of our method over state-of-the-art FS methods in producing high-quality swapped faces.

From Implicit to Explicit feedback: A deep neural network for modeling sequential behaviours and long-short term preferences of online users

In this work, we examine the advantages of using multiple types of behaviour in recommendation systems. Intuitively, each user has to do some implicit actions (e.g., click) before making an explicit decision (e.g., purchase). Previous studies showed that implicit and explicit feedback have different roles for a useful recommendation. However, these studies either exploit implicit and explicit behaviour separately or ignore the semantic of sequential interactions between users and items. In addition, we go from the hypothesis that a user's preference at a time is a combination of long-term and short-term interests. In this paper, we propose some Deep Learning architectures. The first one is Implicit to Explicit (ITE), to exploit users' interests through the sequence of their actions. And two versions of ITE with Bidirectional Encoder Representations from Transformers based (BERT-based) architecture called BERT-ITE and BERT-ITE-Si, which combine users' long- and short-term preferences without and with side information to enhance user representation. The experimental results show that our models outperform previous state-of-the-art ones and also demonstrate our views on the effectiveness of exploiting the implicit to explicit order as well as combining long- and short-term preferences in two large-scale datasets.

Generalization of GANs under Lipschitz continuity and data augmentation

Generative adversarial networks (GANs) have been being widely used in various applications. Arguably, GANs are really complex, and little has been known about their generalization. In this paper, we make a comprehensive analysis about generalization of GANs. We decompose the generalization error into an explicit composition: generator error + discriminator error + optimization error. The first two errors show the capacity of the player's families, are irreducible and optimizer-independent. We then provide both uniform and non-uniform generalization bounds in different scenarios, thanks to our new bridge between Lipschitz continuity and generalization. Our bounds overcome some major limitations of existing ones. In particular, our bounds show that penalizing the zero- and first-order informations of the GAN loss will improve generalization, answering the long mystery of why imposing a Lipschitz constraint can help GANs perform better in practice. Finally, we show why data augmentation penalizes the zero- and first-order informations of the loss, helping the players generalize better, and hence explaining the highly successful use of data augmentation for GANs.

Improving Bayesian Inference in Deep Neural Networks with Variational Structured Dropout

Approximate inference in deep Bayesian networks exhibits a dilemma of how to yield high fidelity posterior approximations while maintaining computational efficiency and scalability. We tackle this challenge by introducing a new variational structured approximation inspired by the interpretation of Dropout training as approximate inference in Bayesian probabilistic models. Concretely, we focus on restrictions of the factorized structure of Dropout posterior which is inflexible to capture rich correlations among weight parameters of the true posterior, and we then propose a novel method called Variational Structured Dropout (VSD) to overcome this limitation. VSD employs an orthogonal transformation to learn a structured representation on the variational Dropout noise and consequently induces statistical dependencies in the approximate posterior. We further gain expressive Bayesian modeling for VSD via proposing a hierarchical Dropout procedure that corresponds to the joint inference in a Bayesian network. Moreover, we can scale up VSD to modern deep convolutional networks in a direct way with a low computational cost. Finally, we conduct extensive experiments on standard benchmarks to demonstrate the effectiveness of VSD over state-of-the-art methods on both predictive accuracy and uncertainty estimation.

Bag of biterms modeling for short texts

Analyzing texts from social media encounters many challenges due to their unique characteristics of shortness, massiveness, and dynamic. Short texts do not provide enough context information, causing the failure of the traditional statistical models. Furthermore, many applications often face with massive and dynamic short texts, causing various computational challenges to the current batch learning algorithms. This paper presents a novel framework, namely Bag of Biterms Modeling (BBM), for modeling massive, dynamic, and short text collections. BBM comprises of two main ingredients: (1) the concept of Bag of Biterms (BoB) for representing documents, and (2) a simple way to help statistical models to include BoB. Our framework can be easily deployed for a large class of probabilistic models, and we demonstrate its usefulness with two well-known models: Latent Dirichlet Allocation (LDA) and Hierarchical Dirichlet Process (HDP). By exploiting both terms (words) and biterms (pairs of words), the major advantages of BBM are: (1) it enhances the length of the documents and makes the context more coherent by emphasizing the word connotation and co-occurrence via Bag of Biterms, (2) it inherits inference and learning algorithms from the primitive to make it straightforward to design online and streaming algorithms for short texts. Extensive experiments suggest that BBM outperforms several state-of-the-art models. We also point out that the BoB representation performs better than the traditional representations (e.g, Bag of Words, tf-idf) even for normal texts.

Graph Convolutional Topic Model for Data Streams

Learning hidden topics in data streams has been paid a great deal of attention by researchers with a lot of proposed methods, but exploiting prior knowledge in general and a knowledge graph in particular has not been taken into adequate consideration in these methods. Prior knowledge that is derived from human knowledge (e.g. Wordnet) or a pre-trained model (e.g.Word2vec) is very valuable and useful to help topic models work better, especially on short texts. However, previous work often ignores this resource, or it can only utilize prior knowledge of a vector form in a simple way. In this paper, we propose a novel graph convolutional topic model (GCTM) which integrates graph convolutional networks (GCN) into a topic model and a learning method which learns the networks and the topic model simultaneously for data streams. In each minibatch, our method not only can exploit an external knowledge graph but also can balance between the external and old knowledge to perform well on new data. We conduct extensive experiments to evaluate our method with both human graph knowledge(Wordnet) and a graph built from pre-trained word embeddings (Word2vec). The experimental results show that our method achieves significantly better performances than the state-of-the-art baselines in terms of probabilistic predictive measure and topic coherence. In particular, our method can work well when dealing with short texts as well as concept drift. The implementation of GCTM is available at https://github.com/bachtranxuan/GCTM.git.