Haokun Chen

Yao Zhang, Haokun Chen, Ahmed Frikha et al. · deepmind, oxford

Visual Question Answering (VQA) is a multi-discipline research task. To produce the right answer, it requires an understanding of the visual content of images, the natural language questions, as well as commonsense reasoning over the information contained in the image and world knowledge. Recently, large-scale Vision-and-Language Pre-trained Models (VLPMs) have been the mainstream approach to VQA tasks due to their superior performance. The standard practice is to fine-tune large-scale VLPMs pre-trained on huge general-domain datasets using the domain-specific VQA datasets. However, in reality, the application domain can change over time, necessitating VLPMs to continually learn and adapt to new domains without forgetting previously acquired knowledge. Most existing continual learning (CL) research concentrates on unimodal tasks, whereas a more practical application scenario, i.e, CL on cross-domain VQA, has not been studied. Motivated by this, we introduce CL-CrossVQA, a rigorous Continual Learning benchmark for Cross-domain Visual Question Answering, through which we conduct extensive experiments on 4 VLPMs, 4 CL approaches, and 5 VQA datasets from different domains. In addition, by probing the forgetting phenomenon of the intermediate layers, we provide insights into how model architecture affects CL performance, why CL approaches can help mitigate forgetting in VLPMs to some extent, and how to design CL approaches suitable for VLPMs in this challenging continual learning environment. To facilitate future work on CL for cross-domain VQA, we will release our datasets and code.

Haokun Chen, Ahmed Frikha, Denis Krompass et al.

Federated Learning (FL) is a decentralized learning paradigm, in which multiple clients collaboratively train deep learning models without centralizing their local data, and hence preserve data privacy. Real-world applications usually involve a distribution shift across the datasets of the different clients, which hurts the generalization ability of the clients to unseen samples from their respective data distributions. In this work, we address the recently proposed feature shift problem where the clients have different feature distributions, while the label distribution is the same. We propose Federated Representation Augmentation (FRAug) to tackle this practical and challenging problem. Our approach generates synthetic client-specific samples in the embedding space to augment the usually small client datasets. For that, we train a shared generative model to fuse the clients knowledge learned from their different feature distributions. This generator synthesizes client-agnostic embeddings, which are then locally transformed into client-specific embeddings by Representation Transformation Networks (RTNets). By transferring knowledge across the clients, the generated embeddings act as a regularizer for the client models and reduce overfitting to the local original datasets, hence improving generalization. Our empirical evaluation on public benchmarks and a real-world medical dataset demonstrates the effectiveness of the proposed method, which substantially outperforms the current state-of-the-art FL methods for non-IID features, including PartialFed and FedBN.

Xu Yang, Yongliang Wu, Mingzhuo Yang et al.

After discovering that Language Models (LMs) can be good in-context few-shot learners, numerous strategies have been proposed to optimize in-context sequence configurations. Recently, researchers in Vision-Language (VL) domains also develop their few-shot learners, while they only use the simplest way, ie., randomly sampling, to configure in-context image-text pairs. In order to explore the effects of varying configurations on VL in-context learning, we devised four strategies for image selection and four for caption assignment to configure in-context image-text pairs for image captioning. Here Image Captioning is used as the case study since it can be seen as the visually-conditioned LM. Our comprehensive experiments yield two counter-intuitive but valuable insights, highlighting the distinct characteristics of VL in-context learning due to multi-modal synergy, as compared to the NLP case. Furthermore, in our exploration of optimal combination strategies, we observed an average performance enhancement of 20.9 of CIDEr scores compared to the baseline. The code is given in https://github.com/yongliang-wu/ExploreCfg.

Jinhe Bi, Danqi Yan, Yifan Wang et al.

Recent Large Reasoning Models significantly improve the reasoning ability of Large Language Models by learning to reason, exhibiting the promising performance in solving complex tasks. LRMs solve tasks that require complex reasoning by explicitly generating reasoning trajectories together with answers. Nevertheless, judging the quality of such an output answer is not easy because only considering the correctness of the answer is not enough and the soundness of the reasoning trajectory part matters as well. Logically, if the soundness of the reasoning part is poor, even if the answer is correct, the confidence of the derived answer should be low. Existing methods did consider jointly assessing the overall output answer by taking into account the reasoning part, however, their capability is still not satisfactory as the causal relationship of the reasoning to the concluded answer cannot properly reflected. In this paper, inspired by classical mechanics, we present a novel approach towards establishing a CoT-Kinetics energy equation. Specifically, our CoT-Kinetics energy equation formulates the token state transformation process, which is regulated by LRM internal transformer layers, as like a particle kinetics dynamics governed in a mechanical field. Our CoT-Kinetics energy assigns a scalar score to evaluate specifically the soundness of the reasoning phase, telling how confident the derived answer could be given the evaluated reasoning. As such, the LRM's overall output quality can be accurately measured, rather than a coarse judgment (e.g., correct or incorrect) anymore.

Jinhe Bi, Yujun Wang, Haokun Chen et al.

Multimodal Large Language Models (MLLMs) have significantly advanced visual tasks by integrating visual representations into large language models (LLMs). The textual modality, inherited from LLMs, equips MLLMs with abilities like instruction following and in-context learning. In contrast, the visual modality enhances performance in downstream tasks by leveraging rich semantic content, spatial information, and grounding capabilities. These intrinsic modalities work synergistically across various visual tasks. Our research initially reveals a persistent imbalance between these modalities, with text often dominating output generation during visual instruction tuning. This imbalance occurs when using both full fine-tuning and parameter-efficient fine-tuning (PEFT) methods. We then found that re-balancing these modalities can significantly reduce the number of trainable parameters required, inspiring a direction for further optimizing visual instruction tuning. We introduce Modality Linear Representation-Steering (MoReS) to achieve the goal. MoReS effectively re-balances the intrinsic modalities throughout the model, where the key idea is to steer visual representations through linear transformations in the visual subspace across each model layer. To validate our solution, we composed LLaVA Steering, a suite of models integrated with the proposed MoReS method. Evaluation results show that the composed LLaVA Steering models require, on average, 500 times fewer trainable parameters than LoRA needs while still achieving comparable performance across three visual benchmarks and eight visual question-answering tasks. Last, we present the LLaVA Steering Factory, an in-house developed platform that enables researchers to quickly customize various MLLMs with component-based architecture for seamlessly integrating state-of-the-art models, and evaluate their intrinsic modality imbalance.

Xiwei Hu, Haokun Chen, Zhongqi Qi et al.

We present DreamPoster, a Text-to-Image generation framework that intelligently synthesizes high-quality posters from user-provided images and text prompts while maintaining content fidelity and supporting flexible resolution and layout outputs. Specifically, DreamPoster is built upon our T2I model, Seedream3.0 to uniformly process different poster generating types. For dataset construction, we propose a systematic data annotation pipeline that precisely annotates textual content and typographic hierarchy information within poster images, while employing comprehensive methodologies to construct paired datasets comprising source materials (e.g., raw graphics/text) and their corresponding final poster outputs. Additionally, we implement a progressive training strategy that enables the model to hierarchically acquire multi-task generation capabilities while maintaining high-quality generation. Evaluations on our testing benchmarks demonstrate DreamPoster's superiority over existing methods, achieving a high usability rate of 88.55\%, compared to GPT-4o (47.56\%) and SeedEdit3.0 (25.96\%). DreamPoster will be online in Jimeng and other Bytedance Apps.

Qi Mao, Haobo Hu, Yujie He et al.

Affective Image Manipulation (AIM) aims to alter visual elements within an image to evoke specific emotional responses from viewers. However, existing AIM approaches rely on rigid \emph{one-to-one} mappings between emotions and visual cues, making them ill-suited for the inherently subjective and diverse ways in which humans perceive and express emotion.To address this, we introduce a novel task setting termed \emph{Diverse AIM (D-AIM)}, aiming to generate multiple visually distinct yet emotionally consistent image edits from a single source image and target emotion. We propose \emph{EmoAgent}, the first multi-agent framework tailored specifically for D-AIM. EmoAgent explicitly decomposes the manipulation process into three specialized phases executed by collaborative agents: a Planning Agent that generates diverse emotional editing strategies, an Editing Agent that precisely executes these strategies, and a Critic Agent that iteratively refines the results to ensure emotional accuracy. This collaborative design empowers EmoAgent to model \emph{one-to-many} emotion-to-visual mappings, enabling semantically diverse and emotionally faithful edits.Extensive quantitative and qualitative evaluations demonstrate that EmoAgent substantially outperforms state-of-the-art approaches in both emotional fidelity and semantic diversity, effectively generating multiple distinct visual edits that convey the same target emotion.