Group imbalance has been a known problem in empirical risk minimization (ERM), where the achieved high average accuracy is accompanied by low accuracy in a minority group. Despite algorithmic efforts to improve the minority group accuracy, a theoretical generalization analysis of ERM on individual groups remains elusive. By formulating the group imbalance problem with the Gaussian Mixture Model, this paper quantifies the impact of individual groups on the sample complexity, the convergence rate, and the average and group-level testing performance. Although our theoretical framework is centered on binary classification using a one-hidden-layer neural network, to the best of our knowledge, we provide the first theoretical analysis of the group-level generalization of ERM in addition to the commonly studied average generalization performance. Sample insights of our theoretical results include that when all group-level co-variance is in the medium regime and all mean are close to zero, the learning performance is most desirable in the sense of a small sample complexity, a fast training rate, and a high average and group-level testing accuracy. Moreover, we show that increasing the fraction of the minority group in the training data does not necessarily improve the generalization performance of the minority group. Our theoretical results are validated on both synthetic and empirical datasets, such as CelebA and CIFAR-10 in image classification.
Micro-action is an imperceptible non-verbal behaviour characterised by low-intensity movement. It offers insights into the feelings and intentions of individuals and is important for human-oriented applications such as emotion recognition and psychological assessment. However, the identification, differentiation, and understanding of micro-actions pose challenges due to the imperceptible and inaccessible nature of these subtle human behaviors in everyday life. In this study, we innovatively collect a new micro-action dataset designated as Micro-action-52 (MA-52), and propose a benchmark named micro-action network (MANet) for micro-action recognition (MAR) task. Uniquely, MA-52 provides the whole-body perspective including gestures, upper- and lower-limb movements, attempting to reveal comprehensive micro-action cues. In detail, MA-52 contains 52 micro-action categories along with seven body part labels, and encompasses a full array of realistic and natural micro-actions, accounting for 205 participants and 22,422 video instances collated from the psychological interviews. Based on the proposed dataset, we assess MANet and other nine prevalent action recognition methods. MANet incorporates squeeze-and excitation (SE) and temporal shift module (TSM) into the ResNet architecture for modeling the spatiotemporal characteristics of micro-actions. Then a joint-embedding loss is designed for semantic matching between video and action labels; the loss is used to better distinguish between visually similar yet distinct micro-action categories. The extended application in emotion recognition has demonstrated one of the important values of our proposed dataset and method. In the future, further exploration of human behaviour, emotion, and psychological assessment will be conducted in depth. The dataset and source code are released at https://github.com/VUT-HFUT/Micro-Action.
Recent works in implicit representations, such as Neural Radiance Fields (NeRF), have advanced the generation of realistic and animatable head avatars from video sequences. These implicit methods are still confronted by visual artifacts and jitters, since the lack of explicit geometric constraints poses a fundamental challenge in accurately modeling complex facial deformations. In this paper, we introduce Dynamic Tetrahedra (DynTet), a novel hybrid representation that encodes explicit dynamic meshes by neural networks to ensure geometric consistency across various motions and viewpoints. DynTet is parameterized by the coordinate-based networks which learn signed distance, deformation, and material texture, anchoring the training data into a predefined tetrahedra grid. Leveraging Marching Tetrahedra, DynTet efficiently decodes textured meshes with a consistent topology, enabling fast rendering through a differentiable rasterizer and supervision via a pixel loss. To enhance training efficiency, we incorporate classical 3D Morphable Models to facilitate geometry learning and define a canonical space for simplifying texture learning. These advantages are readily achievable owing to the effective geometric representation employed in DynTet. Compared with prior works, DynTet demonstrates significant improvements in fidelity, lip synchronization, and real-time performance according to various metrics. Beyond producing stable and visually appealing synthesis videos, our method also outputs the dynamic meshes which is promising to enable many emerging applications.
Transformer-based Single Image Deraining (SID) methods have achieved remarkable success, primarily attributed to their robust capability in capturing long-range interactions. However, we've noticed that current methods handle rain-affected and unaffected regions concurrently, overlooking the disparities between these areas, resulting in confusion between rain streaks and background parts, and inabilities to obtain effective interactions, ultimately resulting in suboptimal deraining outcomes. To address the above issue, we introduce the Region Transformer (Regformer), a novel SID method that underlines the importance of independently processing rain-affected and unaffected regions while considering their combined impact for high-quality image reconstruction. The crux of our method is the innovative Region Transformer Block (RTB), which integrates a Region Masked Attention (RMA) mechanism and a Mixed Gate Forward Block (MGFB). Our RTB is used for attention selection of rain-affected and unaffected regions and local modeling of mixed scales. The RMA generates attention maps tailored to these two regions and their interactions, enabling our model to capture comprehensive features essential for rain removal. To better recover high-frequency textures and capture more local details, we develop the MGFB as a compensation module to complete local mixed scale modeling. Extensive experiments demonstrate that our model reaches state-of-the-art performance, significantly improving the image deraining quality. Our code and trained models are publicly available.
Transformer-based large language models have displayed impressive in-context learning capabilities, where a pre-trained model can handle new tasks without fine-tuning by simply augmenting the query with some input-output examples from that task. Despite the empirical success, the mechanics of how to train a Transformer to achieve ICL and the corresponding ICL capacity is mostly elusive due to the technical challenges of analyzing the nonconvex training problems resulting from the nonlinear self-attention and nonlinear activation in Transformers. To the best of our knowledge, this paper provides the first theoretical analysis of the training dynamics of Transformers with nonlinear self-attention and nonlinear MLP, together with the ICL generalization capability of the resulting model. Focusing on a group of binary classification tasks, we train Transformers using data from a subset of these tasks and quantify the impact of various factors on the ICL generalization performance on the remaining unseen tasks with and without data distribution shifts. We also analyze how different components in the learned Transformers contribute to the ICL performance. Furthermore, we provide the first theoretical analysis of how model pruning affects the ICL performance and prove that proper magnitude-based pruning can have a minimal impact on ICL while reducing inference costs. These theoretical findings are justified through numerical experiments.
While effective in recommendation tasks, collaborative filtering (CF) techniques face the challenge of data sparsity. Researchers have begun leveraging contrastive learning to introduce additional self-supervised signals to address this. However, this approach often unintentionally distances the target user/item from their collaborative neighbors, limiting its efficacy. In response, we propose a solution that treats the collaborative neighbors of the anchor node as positive samples within the final objective loss function. This paper focuses on developing two unique supervised contrastive loss functions that effectively combine supervision signals with contrastive loss. We analyze our proposed loss functions through the gradient lens, demonstrating that different positive samples simultaneously influence updating the anchor node's embeddings. These samples' impact depends on their similarities to the anchor node and the negative samples. Using the graph-based collaborative filtering model as our backbone and following the same data augmentation methods as the existing contrastive learning model SGL, we effectively enhance the performance of the recommendation model. Our proposed Neighborhood-Enhanced Supervised Contrastive Loss (NESCL) model substitutes the contrastive loss function in SGL with our novel loss function, showing marked performance improvement. On three real-world datasets, Yelp2018, Gowalla, and Amazon-Book, our model surpasses the original SGL by 10.09%, 7.09%, and 35.36% on NDCG@20, respectively.
Heterogeneous information networks (HIN) have gained increasing popularity for being able to capture complex relations between nodes of diverse types. Meta-structure was proposed to identify important patterns of relations on HIN, which has been proven effective for extracting rich semantic information and facilitating graph neural networks to learn expressive representations. However, hand-crafted meta-structures pose challenges for scaling up, which draws wide research attention for developing automatic meta-structure search algorithms. Previous efforts concentrate on searching for meta-structures with good empirical prediction performance, overlooking explainability. Thus, they often produce meta-structures prone to overfitting and incomprehensible to humans. To address this, we draw inspiration from the emergent reasoning abilities of large language models (LLMs). We propose a novel REasoning meta-STRUCTure search (ReStruct) framework that integrates LLM reasoning into the evolutionary procedure. ReStruct uses a grammar translator to encode meta-structures into natural language sentences, and leverages the reasoning power of LLMs to evaluate semantically feasible meta-structures. ReStruct also employs performance-oriented evolutionary operations. These two competing forces jointly optimize for semantic explainability and empirical performance of meta-structures. We also design a differential LLM explainer that can produce natural language explanations for the discovered meta-structures, and refine the explanation by reasoning through the search history. Experiments on five datasets demonstrate ReStruct achieve SOTA performance in node classification and link recommendation tasks. Additionally, a survey study involving 73 graduate students shows that the meta-structures and natural language explanations generated by ReStruct are substantially more comprehensible.
Handheld ultrasound devices face usage limitations due to user inexperience and cannot benefit from supervised deep learning without extensive expert annotations. Moreover, the models trained on standard ultrasound device data are constrained by training data distribution and perform poorly when directly applied to handheld device data. In this study, we propose the Training-free Image Style Alignment (TISA) framework to align the style of handheld device data to those of standard devices. The proposed TISA can directly infer handheld device images without extra training and is suited for clinical applications. We show that TISA performs better and more stably in medical detection and segmentation tasks for handheld device data. We further validate TISA as the clinical model for automatic measurements of spinal curvature and carotid intima-media thickness. The automatic measurements agree well with manual measurements made by human experts and the measurement errors remain within clinically acceptable ranges. We demonstrate the potential for TISA to facilitate automatic diagnosis on handheld ultrasound devices and expedite their eventual widespread use.
Human mobility behaviours are closely linked to various important societal problems such as traffic congestion, and epidemic control. However, collecting mobility data can be prohibitively expensive and involves serious privacy issues, posing a pressing need for high-quality generative mobility models. Previous efforts focus on learning the behaviour distribution from training samples, and generate new mobility data by sampling the learned distributions. They cannot effectively capture the coherent intentions that drive mobility behavior, leading to low sample efficiency and semantic-awareness. Inspired by the emergent reasoning ability in LLMs, we propose a radical perspective shift that reformulates mobility generation as a commonsense reasoning problem. In this paper, we design a novel Mobility Generation as Reasoning (MobiGeaR) framework that prompts LLM to recursively generate mobility behaviour. Specifically, we design a context-aware chain-of-thoughts prompting technique to align LLMs with context-aware mobility behaviour by few-shot in-context learning. Besides, MobiGeaR employ a divide-and-coordinate mechanism to exploit the synergistic effect between LLM reasoning and mechanistic gravity model. It leverages the step-by-step LLM reasoning to recursively generate a temporal template of activity intentions, which are then mapped to physical locations with a mechanistic gravity model. Experiments on two real-world datasets show MobiGeaR achieves state-of-the-art performance across all metrics, and substantially reduces the size of training samples at the same time. Besides, MobiGeaR also significantly improves the semantic-awareness of mobility generation by improving the intention accuracy by 62.23% and the generated mobility data is proven effective in boosting the performance of downstream applications. The implementation of our approach is available in the paper.
Visual prompt tuning (VPT) is a promising solution incorporating learnable prompt tokens to customize pre-trained models for downstream tasks. However, VPT and its variants often encounter challenges like prompt initialization, prompt length, and subpar performance in self-supervised pretraining, hindering successful contextual adaptation. This study commences by exploring the correlation evolvement between prompts and patch tokens during proficient training. Inspired by the observation that the prompt tokens tend to share high mutual information with patch tokens, we propose initializing prompts with downstream token prototypes. The strategic initialization, a stand-in for the previous initialization, substantially improves performance in fine-tuning. To refine further, we optimize token construction with a streamlined pipeline that maintains excellent performance with almost no increase in computational expenses compared to VPT. Exhaustive experiments show our proposed approach outperforms existing methods by a remarkable margin. For instance, it surpasses full fine-tuning in 19 out of 24 tasks, using less than 0.4% of learnable parameters on the FGVC and VTAB-1K benchmarks. Notably, our method significantly advances the adaptation for self-supervised pretraining, achieving impressive task performance gains of at least 10% to 30%. Besides, the experimental results demonstrate the proposed SPT is robust to prompt lengths and scales well with model capacity and training data size. We finally provide an insightful exploration into the amount of target data facilitating the adaptation of pre-trained models to downstream tasks.