Yue Zhang

Hossein Rajaby Faghihi, Aliakbar Nafar, Chen Zheng et al. · berkeley

Recent research has shown that integrating domain knowledge into deep learning architectures is effective -- it helps reduce the amount of required data, improves the accuracy of the models' decisions, and improves the interpretability of models. However, the research community is missing a convened benchmark for systematically evaluating knowledge integration methods. In this work, we create a benchmark that is a collection of nine tasks in the domains of natural language processing and computer vision. In all cases, we model external knowledge as constraints, specify the sources of the constraints for each task, and implement various models that use these constraints. We report the results of these models using a new set of extended evaluation criteria in addition to the task performances for a more in-depth analysis. This effort provides a framework for a more comprehensive and systematic comparison of constraint integration techniques and for identifying related research challenges. It will facilitate further research for alleviating some problems of state-of-the-art neural models.

Yue Zhang, Parisa Kordjamshidi

Language understanding is essential for the navigation agent to follow instructions. We observe two kinds of issues in the instructions that can make the navigation task challenging: 1. The mentioned landmarks are not recognizable by the navigation agent due to the different vision abilities of the instructor and the modeled agent. 2. The mentioned landmarks are applicable to multiple targets, thus not distinctive for selecting the target among the candidate viewpoints. To deal with these issues, we design a translator module for the navigation agent to convert the original instructions into easy-to-follow sub-instruction representations at each step. The translator needs to focus on the recognizable and distinctive landmarks based on the agent's visual abilities and the observed visual environment. To achieve this goal, we create a new synthetic sub-instruction dataset and design specific tasks to train the translator and the navigation agent. We evaluate our approach on Room2Room~(R2R), Room4room~(R4R), and Room2Room Last (R2R-Last) datasets and achieve state-of-the-art results on multiple benchmarks.

Yue Zhang, Parisa Kordjamshidi

Understanding spatial and visual information is essential for a navigation agent who follows natural language instructions. The current Transformer-based VLN agents entangle the orientation and vision information, which limits the gain from the learning of each information source. In this paper, we design a neural agent with explicit Orientation and Vision modules. Those modules learn to ground spatial information and landmark mentions in the instructions to the visual environment more effectively. To strengthen the spatial reasoning and visual perception of the agent, we design specific pre-training tasks to feed and better utilize the corresponding modules in our final navigation model. We evaluate our approach on both Room2room (R2R) and Room4room (R4R) datasets and achieve the state of the art results on both benchmarks.

Hongmin Cai, Fei Qi, Junyu Li et al.

Conventional clustering methods based on pairwise affinity usually suffer from the concentration effect while processing huge dimensional features yet low sample sizes data, resulting in inaccuracy to encode the sample proximity and suboptimal performance in clustering. To address this issue, we propose a unified tensor clustering method (UTC) that characterizes sample proximity using multiple samples' affinity, thereby supplementing rich spatial sample distributions to boost clustering. Specifically, we find that the triadic tensor affinity can be constructed via the Khari-Rao product of two affinity matrices. Furthermore, our early work shows that the fourth-order tensor affinity is defined by the Kronecker product. Therefore, we utilize arithmetical products, Khatri-Rao and Kronecker products, to mathematically integrate different orders of affinity into a unified tensor clustering framework. Thus, the UTC jointly learns a joint low-dimensional embedding to combine various orders. Finally, a numerical scheme is designed to solve the problem. Experiments on synthetic datasets and real-world datasets demonstrate that 1) the usage of high-order tensor affinity could provide a supplementary characterization of sample proximity to the popular affinity matrix; 2) the proposed method of UTC is affirmed to enhance clustering by exploiting different order affinities when processing high-dimensional data.

Yue Zhang, Parisa Kordjamshidi

The existing methods for Vision and Language Navigation in the Continuous Environment (VLN-CE) commonly incorporate a waypoint predictor to discretize the environment. This simplifies the navigation actions into a view selection task and improves navigation performance significantly compared to direct training using low-level actions. However, the VLN-CE agents are still far from the real robots since there are gaps between their visual perception and executed actions. First, VLN-CE agents that discretize the visual environment are primarily trained with high-level view selection, which causes them to ignore crucial spatial reasoning within the low-level action movements. Second, in these models, the existing waypoint predictors neglect object semantics and their attributes related to passibility, which can be informative in indicating the feasibility of actions. To address these two issues, we introduce a low-level action decoder jointly trained with high-level action prediction, enabling the current VLN agent to learn and ground the selected visual view to the low-level controls. Moreover, we enhance the current waypoint predictor by utilizing visual representations containing rich semantic information and explicitly masking obstacles based on humans' prior knowledge about the feasibility of actions. Empirically, our agent can improve navigation performance metrics compared to the strong baselines on both high-level and low-level actions.

Ziqi Zhang, Yile Wang, Yue Zhang et al.

Pre-training has been a useful method for learning implicit transferable knowledge and it shows the benefit of offering complementary features across different modalities. Recent work mainly focuses on the modalities such as image and text, for example, studies show that visual features learned from images can help visual-grounded language understanding. In this paper, we consider investigating the potential connection between offline reinforcement learning (RL) and language modeling (LM). Intuitively, RL and LM are similar in predicting the next states based on the current and previous states, which rely on both local and long-range dependency across states. To validate such an assumption, we pre-trained different offline RL tasks using Transformer and then evaluate these models on various language-related tasks. Experimental results show that our RL pre-trained models can give close performance compared with the models using the LM training objective, showing that there exist common useful features across these two modalities. To further explore the potential relationship, we investigate some factors such as Markov property and the sequential nature of RL trajectory.

Qiji Zhou, Yifan Gong, Guangsheng Bao et al.

Counterfactual reasoning is crucial for robust video understanding but remains underexplored in existing multimodal benchmarks. In this paper, we introduce \textbf{COVER} (\textbf{\underline{CO}}unterfactual \textbf{\underline{V}}id\textbf{\underline{E}}o \textbf{\underline{R}}easoning), a multidimensional multimodal benchmark that systematically evaluates MLLMs across the abstract-concrete and perception-cognition dimensions. Beyond prior multimodal benchmarks, COVER decomposes complex queries into structured sub-questions, enabling fine-grained reasoning analysis. Experiments on commercial and open-source models reveal a strong correlation between sub-question accuracy and counterfactual reasoning performance, highlighting the role of structured inference in video understanding. Furthermore, our results suggest a key insight: enhancing the reasoning capability of models is essential for improving the robustness of video understanding. COVER establishes a new standard for assessing MLLMs' logical reasoning abilities in dynamic environments. Our work is available at https://github.com/gongyifan-hash/COVER-Benchmark.

Hanxu Hu, Hongyuan Lu, Huajian Zhang et al.

In this paper, we take the initiative to investigate the performance of LLMs on complex planning tasks that require LLMs to understand a virtual spatial environment simulated via natural language and act correspondingly in text. We propose a benchmark named Natural Language Planning and Action (Natala) composed of a set of novel tasks: Brick World, NLVR-based Manipulations, and Natural Language Navigation. We found that current popular LLMs such as ChatGPT still lack abilities in complex planning. This arises a question -- do the LLMs have a good understanding of the environments described in natural language, or maybe other alternatives such as symbolic representations are neater and hence better to be understood by LLMs? To this end, we propose a novel method called CoS (Chain-of-Symbol Prompting) that represents the complex environments with condensed symbolic spatial representations during the chained intermediate thinking steps. CoS is easy to use and does not need additional training on LLMs. Extensive experiments indicate that CoS clearly surpasses the performance of the Chain-of-Thought (CoT) Prompting in all three planning tasks with even fewer tokens used in the inputs compared with CoT on ChatGPT and InstructGPT. The performance gain is strong, by up to 60.8% accuracy (from 31.8% to 92.6%) on Brick World for ChatGPT. CoS also reduces the number of tokens in the prompt obviously, by up to 65.8% of the tokens (from 407 to 139) for the intermediate steps from demonstrations on Brick World. Code and data available at: https://github.com/hanxuhu/chain-of-symbol-planning

Qiji Zhou, Ruochen Zhou, Zike Hu et al.

Recent advancements in Chain-of-Thought (CoT) and related rationale-based works have significantly improved the performance of Large Language Models (LLMs) in complex reasoning tasks. With the evolution of Multimodal Large Language Models (MLLMs), enhancing their capability to tackle complex multimodal reasoning problems is a crucial frontier. However, incorporating multimodal rationales in CoT has yet to be thoroughly investigated. We propose the Image-of-Thought (IoT) prompting method, which helps MLLMs to extract visual rationales step-by-step. Specifically, IoT prompting can automatically design critical visual information extraction operations based on the input images and questions. Each step of visual information refinement identifies specific visual rationales that support answers to complex visual reasoning questions. Beyond the textual CoT, IoT simultaneously utilizes visual and textual rationales to help MLLMs understand complex multimodal information. IoT prompting has improved zero-shot visual reasoning performance across various visual understanding tasks in different MLLMs. Moreover, the step-by-step visual feature explanations generated by IoT prompting elucidate the visual reasoning process, aiding in analyzing the cognitive processes of large multimodal models

Yue Zhang, Quan Guo, Parisa Kordjamshidi

Existing work on vision and language navigation mainly relies on navigation-related losses to establish the connection between vision and language modalities, neglecting aspects of helping the navigation agent build a deep understanding of the visual environment. In our work, we provide indirect supervision to the navigation agent through a hint generator that provides detailed visual descriptions. The hint generator assists the navigation agent in developing a global understanding of the visual environment. It directs the agent's attention toward related navigation details, including the relevant sub-instruction, potential challenges in recognition and ambiguities in grounding, and the targeted viewpoint description. To train the hint generator, we construct a synthetic dataset based on landmarks in the instructions and visible and distinctive objects in the visual environment. We evaluate our method on the R2R and R4R datasets and achieve state-of-the-art on several metrics. The experimental results demonstrate that generating hints not only enhances the navigation performance but also helps improve the interpretability of the agent's actions.

Yue Zhang, Liqiang Jing, Vibhav Gogate

We introduce a new task called Defeasible Visual Entailment (DVE), where the goal is to allow the modification of the entailment relationship between an image premise and a text hypothesis based on an additional update. While this concept is well-established in Natural Language Inference, it remains unexplored in visual entailment. At a high level, DVE enables models to refine their initial interpretations, leading to improved accuracy and reliability in various applications such as detecting misleading information in images, enhancing visual question answering, and refining decision-making processes in autonomous systems. Existing metrics do not adequately capture the change in the entailment relationship brought by updates. To address this, we propose a novel inference-aware evaluator designed to capture changes in entailment strength induced by updates, using pairwise contrastive learning and categorical information learning. Additionally, we introduce a reward-driven update optimization method to further enhance the quality of updates generated by multimodal models. Experimental results demonstrate the effectiveness of our proposed evaluator and optimization method.

Yue Zhang, Hehe Fan, Yi Yang

To bridge the gap between vision and language modalities, Multimodal Large Language Models (MLLMs) usually learn an adapter that converts visual inputs to understandable tokens for Large Language Models (LLMs). However, most adapters generate consistent visual tokens, regardless of the specific objects of interest mentioned in the prompt. Since these adapters distribute equal attention to every detail in the image and focus on the entire scene, they may increase the cognitive load for LLMs, particularly when processing complex scenes. To alleviate this problem, we propose prompt-aware adapters. These adapters are designed with the capability to dynamically embed visual inputs based on the specific focus of the prompt. Specifically, prompt-aware adapters utilize both global and local textual features to capture the most relevant visual clues from the prompt at both coarse and fine granularity levels. This approach significantly enhances the ability of LLMs to understand and interpret visual content. Experiments on various visual question answering tasks, such as counting and position reasoning, demonstrate the effectiveness of prompt-aware adapters.

Tianyi Ma, Yue Zhang, Zehao Wang et al.

Vision-and-Language Navigation (VLN) poses significant challenges for agents to interpret natural language instructions and navigate complex 3D environments. While recent progress has been driven by large-scale pre-training and data augmentation, current methods still struggle to generalize to unseen scenarios, particularly when complex spatial and temporal reasoning is required. In this work, we propose SkillNav, a modular framework that introduces structured, skill-based reasoning into Transformer-based VLN agents. Our method decomposes navigation into a set of interpretable atomic skills (e.g., Vertical Movement, Area and Region Identification, Stop and Pause), each handled by a specialized agent. To support targeted skill training without manual data annotation, we construct a synthetic dataset pipeline that generates diverse, linguistically natural, skill-specific instruction-trajectory pairs. We then introduce a novel training-free Vision-Language Model (VLM)-based router, which dynamically selects the most suitable agent at each time step by aligning sub-goals with visual observations and historical actions. SkillNav obtains competitive results on commonly used benchmarks and establishes state-of-the-art generalization to the GSA-R2R, a benchmark with novel instruction styles and unseen environments.

Yue Zhang, Jilei Sun, Yunhui Guo et al.

Video Large Multimodal Models (VLMMs) have made impressive strides in understanding video content, but they often struggle with abstract and adaptive reasoning-the ability to revise their interpretations when new information emerges. In reality, conclusions are rarely set in stone; additional context can strengthen or weaken an initial inference. To address this, we introduce Defeasible Video Entailment (DVidE), a new task that challenges models to think like doubters, constantly updating their reasoning based on evolving evidence. In DVidE, given a video premise and a textual hypothesis, models must determine whether a new update strengthens or weakens the hypothesis (classification version) or generate a coherent update that modifies the entailment relationship (generation version). For solving the classification task, we propose the Chain of Counterfactual Thought framework, utilizing counterfactual reasoning, ASR-enhanced video content, and rationale refinement to reduce inference bias. For the generation task, we develop a framework that combines ASR output with a Large Language Model (LLM) to produce coherent, contextually relevant updates aligned with the intended strengthener or weakener goals. Additionally, we introduce a novel benchmark dataset, with strengthener/weakener annotations and an LLM-based evaluation metric specifically designed for assessing generative performance. Experimental results demonstrate significant improvements, highlighting our proposed method in enhancing dynamic reasoning capabilities of VLMMs.

Yue Zhang, Quan Guo, Parisa Kordjamshidi

We deal with the navigation problem where the agent follows natural language instructions while observing the environment. Focusing on language understanding, we show the importance of spatial semantics in grounding navigation instructions into visual perceptions. We propose a neural agent that uses the elements of spatial configurations and investigate their influence on the navigation agent's reasoning ability. Moreover, we model the sequential execution order and align visual objects with spatial configurations in the instruction. Our neural agent improves strong baselines on the seen environments and shows competitive performance on the unseen environments. Additionally, the experimental results demonstrate that explicit modeling of spatial semantic elements in the instructions can improve the grounding and spatial reasoning of the model.

Yue Zhang, Shun Miao, Tommaso Mansi et al.

Automatic parsing of anatomical objects in X-ray images is critical to many clinical applications in particular towards image-guided invention and workflow automation. Existing deep network models require a large amount of labeled data. However, obtaining accurate pixel-wise labeling in X-ray images relies heavily on skilled clinicians due to the large overlaps of anatomy and the complex texture patterns. On the other hand, organs in 3D CT scans preserve clearer structures as well as sharper boundaries and thus can be easily delineated. In this paper, we propose a novel model framework for learning automatic X-ray image parsing from labeled CT scans. Specifically, a Dense Image-to-Image network (DI2I) for multi-organ segmentation is first trained on X-ray like Digitally Reconstructed Radiographs (DRRs) rendered from 3D CT volumes. Then we introduce a Task Driven Generative Adversarial Network (TD-GAN) architecture to achieve simultaneous style transfer and parsing for unseen real X-ray images. TD-GAN consists of a modified cycle-GAN substructure for pixel-to-pixel translation between DRRs and X-ray images and an added module leveraging the pre-trained DI2I to enforce segmentation consistency. The TD-GAN framework is general and can be easily adapted to other learning tasks. In the numerical experiments, we validate the proposed model on 815 DRRs and 153 topograms. While the vanilla DI2I without any adaptation fails completely on segmenting the topograms, the proposed model does not require any topogram labels and is able to provide a promising average dice of 85% which achieves the same level accuracy of supervised training (88%).