Diverse video captioning aims to generate a set of sentences to describe the given video in various aspects. Mainstream methods are trained with independent pairs of a video and a caption from its ground-truth set without exploiting the intra-set relationship, resulting in low diversity of generated captions. Different from them, we formulate diverse captioning into a semantic-concept-guided set prediction (SCG-SP) problem by fitting the predicted caption set to the ground-truth set, where the set-level relationship is fully captured. Specifically, our set prediction consists of two synergistic tasks, i.e., caption generation and an auxiliary task of concept combination prediction providing extra semantic supervision. Each caption in the set is attached to a concept combination indicating the primary semantic content of the caption and facilitating element alignment in set prediction. Furthermore, we apply a diversity regularization term on concepts to encourage the model to generate semantically diverse captions with various concept combinations. These two tasks share multiple semantics-specific encodings as input, which are obtained by iterative interaction between visual features and conceptual queries. The correspondence between the generated captions and specific concept combinations further guarantees the interpretability of our model. Extensive experiments on benchmark datasets show that the proposed SCG-SP achieves state-of-the-art (SOTA) performance under both relevance and diversity metrics.
This paper introduces "Shai" a 10B level large language model specifically designed for the asset management industry, built upon an open-source foundational model. With continuous pre-training and fine-tuning using a targeted corpus, Shai demonstrates enhanced performance in tasks relevant to its domain, outperforming baseline models. Our research includes the development of an innovative evaluation framework, which integrates professional qualification exams, tailored tasks, open-ended question answering, and safety assessments, to comprehensively assess Shai's capabilities. Furthermore, we discuss the challenges and implications of utilizing large language models like GPT-4 for performance assessment in asset management, suggesting a combination of automated evaluation and human judgment. Shai's development, showcasing the potential and versatility of 10B-level large language models in the financial sector with significant performance and modest computational requirements, hopes to provide practical insights and methodologies to assist industry peers in their similar endeavors.
Introducing BERT into cross-modal settings raises difficulties in its optimization for handling multiple modalities. Both the BERT architecture and training objective need to be adapted to incorporate and model information from different modalities. In this paper, we address these challenges by exploring the implicit semantic and geometric correlations between 2D and 3D data of the same objects/scenes. We propose a new cross-modal BERT-style self-supervised learning paradigm, called Cross-BERT. To facilitate pretraining for irregular and sparse point clouds, we design two self-supervised tasks to boost cross-modal interaction. The first task, referred to as Point-Image Alignment, aims to align features between unimodal and cross-modal representations to capture the correspondences between the 2D and 3D modalities. The second task, termed Masked Cross-modal Modeling, further improves mask modeling of BERT by incorporating high-dimensional semantic information obtained by cross-modal interaction. By performing cross-modal interaction, Cross-BERT can smoothly reconstruct the masked tokens during pretraining, leading to notable performance enhancements for downstream tasks. Through empirical evaluation, we demonstrate that Cross-BERT outperforms existing state-of-the-art methods in 3D downstream applications. Our work highlights the effectiveness of leveraging cross-modal 2D knowledge to strengthen 3D point cloud representation and the transferable capability of BERT across modalities.
Despite the effectiveness in improving the robustness of neural networks, adversarial training has suffered from the natural accuracy degradation problem, i.e., accuracy on natural samples has reduced significantly. In this study, we reveal that natural accuracy degradation is highly related to the disruption of the natural sample topology in the representation space by quantitative and qualitative experiments. Based on this observation, we propose Topology-pReserving Adversarial traINing (TRAIN) to alleviate the problem by preserving the topology structure of natural samples from a standard model trained only on natural samples during adversarial training. As an additional regularization, our method can easily be combined with various popular adversarial training algorithms in a plug-and-play manner, taking advantage of both sides. Extensive experiments on CIFAR-10, CIFAR-100, and Tiny ImageNet show that our proposed method achieves consistent and significant improvements over various strong baselines in most cases. Specifically, without additional data, our proposed method achieves up to 8.78% improvement in natural accuracy and 4.50% improvement in robust accuracy.
Despite the remarkable advances that have been made in continual learning, the adversarial vulnerability of such methods has not been fully discussed. We delve into the adversarial robustness of memory-based continual learning algorithms and observe limited robustness improvement by directly applying adversarial training techniques. Preliminary studies reveal the twin challenges for building adversarial robust continual learners: accelerated forgetting in continual learning and gradient obfuscation in adversarial robustness. In this study, we put forward a novel adversarial robust memory-based continual learner that adjusts data logits to mitigate the forgetting of pasts caused by adversarial samples. Furthermore, we devise a gradient-based data selection mechanism to overcome the gradient obfuscation caused by limited stored data. The proposed approach can widely integrate with existing memory-based continual learning as well as adversarial training algorithms in a plug-and-play way. Extensive experiments on Split-CIFAR10/100 and Split-Tiny-ImageNet demonstrate the effectiveness of our approach, achieving up to 8.13% higher accuracy for adversarial data.
Generating vivid and emotional 3D co-speech gestures is crucial for virtual avatar animation in human-machine interaction applications. While the existing methods enable generating the gestures to follow a single emotion label, they overlook that long gesture sequence modeling with emotion transition is more practical in real scenes. In addition, the lack of large-scale available datasets with emotional transition speech and corresponding 3D human gestures also limits the addressing of this task. To fulfill this goal, we first incorporate the ChatGPT-4 and an audio inpainting approach to construct the high-fidelity emotion transition human speeches. Considering obtaining the realistic 3D pose annotations corresponding to the dynamically inpainted emotion transition audio is extremely difficult, we propose a novel weakly supervised training strategy to encourage authority gesture transitions. Specifically, to enhance the coordination of transition gestures w.r.t different emotional ones, we model the temporal association representation between two different emotional gesture sequences as style guidance and infuse it into the transition generation. We further devise an emotion mixture mechanism that provides weak supervision based on a learnable mixed emotion label for transition gestures. Last, we present a keyframe sampler to supply effective initial posture cues in long sequences, enabling us to generate diverse gestures. Extensive experiments demonstrate that our method outperforms the state-of-the-art models constructed by adapting single emotion-conditioned counterparts on our newly defined emotion transition task and datasets.
Vision-language models (VLMs) have recently shown promising results in traditional downstream tasks. Evaluation studies have emerged to assess their abilities, with the majority focusing on the third-person perspective, and only a few addressing specific tasks from the first-person perspective. However, the capability of VLMs to "think" from a first-person perspective, a crucial attribute for advancing autonomous agents and robotics, remains largely unexplored. To bridge this research gap, we introduce EgoThink, a novel visual question-answering benchmark that encompasses six core capabilities with twelve detailed dimensions. The benchmark is constructed using selected clips from egocentric videos, with manually annotated question-answer pairs containing first-person information. To comprehensively assess VLMs, we evaluate eighteen popular VLMs on EgoThink. Moreover, given the open-ended format of the answers, we use GPT-4 as the automatic judge to compute single-answer grading. Experimental results indicate that although GPT-4V leads in numerous dimensions, all evaluated VLMs still possess considerable potential for improvement in first-person perspective tasks. Meanwhile, enlarging the number of trainable parameters has the most significant impact on model performance on EgoThink. In conclusion, EgoThink serves as a valuable addition to existing evaluation benchmarks for VLMs, providing an indispensable resource for future research in the realm of embodied artificial intelligence and robotics.
Large Language Models (LLMs) demonstrate impressive reasoning ability and the maintenance of world knowledge not only in natural language tasks, but also in some vision-language tasks such as open-domain knowledge-based visual question answering (OK-VQA). As images are invisible to LLMs, researchers convert images to text to engage LLMs into the visual question reasoning procedure. This leads to discrepancies between images and their textual representations presented to LLMs, which consequently impedes final reasoning performance. To fill the information gap and better leverage the reasoning capability, we design a framework that enables LLMs to proactively ask relevant questions to unveil more details in the image, along with filters for refining the generated information. We validate our idea on OK-VQA and A-OKVQA. Our method continuously boosts the performance of baselines methods by an average gain of 2.15% on OK-VQA, and achieves consistent improvements across different LLMs.
Among the array of neural network architectures, the Vision Transformer (ViT) stands out as a prominent choice, acclaimed for its exceptional expressiveness and consistent high performance in various vision applications. Recently, the emerging Spiking ViT approach has endeavored to harness spiking neurons, paving the way for a more brain-inspired transformer architecture that thrives in ultra-low power operations on dedicated neuromorphic hardware. Nevertheless, this approach remains confined to spatial self-attention and doesn't fully unlock the potential of spiking neural networks. We introduce DISTA, a Denoising Spiking Transformer with Intrinsic Plasticity and SpatioTemporal Attention, designed to maximize the spatiotemporal computational prowess of spiking neurons, particularly for vision applications. DISTA explores two types of spatiotemporal attentions: intrinsic neuron-level attention and network-level attention with explicit memory. Additionally, DISTA incorporates an efficient nonlinear denoising mechanism to quell the noise inherent in computed spatiotemporal attention maps, thereby resulting in further performance gains. Our DISTA transformer undergoes joint training involving synaptic plasticity (i.e., weight tuning) and intrinsic plasticity (i.e., membrane time constant tuning) and delivers state-of-the-art performances across several static image and dynamic neuromorphic datasets. With only 6 time steps, DISTA achieves remarkable top-1 accuracy on CIFAR10 (96.26%) and CIFAR100 (79.15%), as well as 79.1% on CIFAR10-DVS using 10 time steps.
Large Language Models (LLMs) have showcased impressive performance. However, due to their inability to capture relationships among samples, these frozen LLMs inevitably keep repeating similar mistakes. In this work, we propose our Tuning-free Rule Accumulation (TRAN) framework, which guides LLMs in improving their performance by learning from previous mistakes. Considering data arrives sequentially, LLMs gradually accumulate rules from incorrect cases, forming a rule collection. These rules are then utilized by the LLMs to avoid making similar mistakes when processing subsequent inputs. Moreover, the rules remain independent of the primary prompts, seamlessly complementing prompt design strategies. Experimentally, we show that TRAN improves over recent baselines by a large margin.