Self-correction is essential for solving complex reasoning problems in vision-language models (VLMs). However, existing reinforcement learning (RL) methods struggle to learn it, as effective self-correction behaviors emerge only rarely, making learning signals extremely sparse. To address this challenge, we propose correction-specific rollouts (Octopus), an RL rollout augmentation framework that synthesizes dense self-correction examples by recombining existing rollouts. This augmentation simultaneously improves sample efficiency due to rollout reuse and stabilizes RL optimization through balanced supervision. Furthermore, we introduce a response-masking strategy that decouples self-correction from direct reasoning, avoiding signal conflicts and enabling both behaviors to be learned effectively. Building on this, we introduce Octopus-8B, a reasoning VLM with controllable self-correction capability. Across 7 benchmarks, it achieves SoTA performance among open-source VLMs, outperforming the best RLVR baseline by 1.0 score while requiring only $0.72\times$ training time per step.
Exploiting Generalization in Offline Reinforcement Learning via Unseen State Augmentations
Georgia Institute of Technology（佐治亚理工学院）；Amazon Alexa AI（亚马逊Alexa人工智能）；Allen Institute for Artificial Intelligence（艾伦人工智能研究所）；University of Southern California（南加州大学）
Model-Based Offline Reinforcement Learning with Adversarial Data Augmentation
National Key Laboratory for Novel Software Technology, Nanjing University（南京大学现代软件技术国家重点实验室）；City University of Hong Kong（香港城市大学）；School of Computer Science, Nanjing University of Posts and Telecommunications（南京邮电大学计算机学院）；Department of Computer Science, University of Liverpool（利物浦大学计算机系）
Prioritized Rollouts for Efficient World Model-based Vision-Language-Action Policy Optimization
National Key Laboratory for Novel Software Technology, Nanjing University（南京大学计算机软件新技术全国重点实验室）；School of Artificial Intelligence, Nanjing University（南京大学人工智能学院）；Cirquar Technologies；Mila – Quebec AI Institute（Mila – 魁北克人工智能研究所）；Université de Montréal（蒙特利尔大学）
While agents trained by Reinforcement Learning (RL) can solve increasingly challenging tasks directly from visual observations, generalizing learned skills to novel environments remains very challenging. Extensive use of data augmentation is a promising technique for improving generalization in RL, but it is often found to decrease sample efficiency and can even lead to divergence. In this paper, we investigate causes of instability when using data augmentation in common off-policy RL algorithms. We identify two problems, both rooted in high-variance Q-targets. Based on our findings, we propos…
Florida Institute for Human and Machine Cognition (IHMC)（佛罗里达人机认知研究所（IHMC））
Despite advancements in deep reinforcement learning algorithms, developing an effective exploration strategy is still an open problem. Most existing exploration strategies either are based on simple heuristics, or require the model of the environment, or train additional deep neural networks to generate imagination-augmented paths. In this paper, a revolutionary algorithm, called Policy Augmentation, is introduced. Policy Augmentation is based on a newly developed inductive matrix completion method. The proposed algorithm augments the values of unexplored state-action pairs, helping the agent …
Learning from Hindsight for VLA Reinforcement Learning
Massachusetts Institute of Technology（麻省理工学院）；MIT-IBM Computing Research Lab（麻省理工学院-IBM计算研究实验室）；Stanford University（斯坦福大学）；University of California, San Diego（加利福尼亚大学圣地亚哥分校）
Institute of Information Engineering, CAS（中国科学院信息工程研究所）；Amazon Web Services（亚马逊云科技）
Data augmentation is a necessity to enhance data efficiency in deep learning. For vision-language pre-training, data is only augmented either for images or for text in previous works. In this paper, we present MixGen: a joint data augmentation for vision-language representation learning to further improve data efficiency. It generates new image-text pairs with semantic relationships preserved by interpolating images and concatenating text. It's simple, and can be plug-and-played into existing pipelines. We evaluate MixGen on four architectures, including CLIP, ViLT, ALBEF and TCL, across five …
Can Vision-Language Models Stay Helpful When Facing Implicit Risks? Intent-Privilege OPSD for Efficient Safety-Helpfulness Alignment
Southern University of Science and Technology（南方科技大学）；Sun Yat-sen University（中山大学）；Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ)（广东省人工智能与数字经济实验室（深圳））；University of Electronic Science and Technology of China（电子科技大学）；Hangzhou Dianzi University（杭州电子科技大学）；Shenzhen MSU-BIT University（深圳北理莫斯科大学）；GigaAI（极佳科技）；The Chinese University of Hong Kong (Shenzhen)（香港中文大学（深圳））
AGC: Adaptive Geodesic Correction for Adversarial Robustness on Vision-Language Models
NLPR & MAIS, Institute of Automation, Chinese Academy of Sciences（自动化研究所国家工程研究中心与人工智能院，中国科学院）；School of Computer Science and Engineering, Central South University（中南大学计算机科学与工程学院）；University of Chinese Academy of Sciences（中国科学院大学）
MetaTPT: Meta Test-time Prompt Tuning for Vision-Language Models
UCAS-Terminus AI Lab, University of Chinese Academic of Sciences（中国科学院大学Terminus AI实验室，中国科学院大学）；AIM Lab, University of Amsterdam（阿姆斯特丹大学AIM实验室）；Jarvis Research Center, Tencent Youtu Lab（腾讯优图实验室 Jarvis 研究中心）；Central Research Institue, United Imaging Healthcare Co., Ltd（联合影像医疗科技有限公司中央研究所）
University of Maryland（马里兰大学）；New York University Center for Data Science（纽约大学数据科学中心）
Existing techniques for model inversion typically rely on hard-to-tune regularizers, such as total variation or feature regularization, which must be individually calibrated for each network in order to produce adequate images. In this work, we introduce Plug-In Inversion, which relies on a simple set of augmentations and does not require excessive hyper-parameter tuning. Under our proposed augmentation-based scheme, the same set of augmentation hyper-parameters can be used for inverting a wide range of image classification models, regardless of input dimensions or the architecture. We illustr…
Vision Transformers (ViTs) and MLPs signal further efforts on replacing hand-wired features or inductive biases with general-purpose neural architectures. Existing works empower the models by massive data, such as large-scale pre-training and/or repeated strong data augmentations, and still report optimization-related problems (e.g., sensitivity to initialization and learning rates). Hence, this paper investigates ViTs and MLP-Mixers from the lens of loss geometry, intending to improve the models' data efficiency at training and generalization at inference. Visualization and Hessian reveal ext…
We investigate the robustness of vision transformers (ViTs) through the lens of their special patch-based architectural structure, i.e., they process an image as a sequence of image patches. We find that ViTs are surprisingly insensitive to patch-based transformations, even when the transformation largely destroys the original semantics and makes the image unrecognizable by humans. This indicates that ViTs heavily use features that survived such transformations but are generally not indicative of the semantic class to humans. Further investigations show that these features are useful but non-r…
Institute of Cognitive Science（认知科学研究所）；University of Osnabrück（奥斯纳布吕克大学）
Interdisciplinary research is often at the core of scientific progress. This dissertation explores some advantageous synergies between machine learning, cognitive science and neuroscience. In particular, this thesis focuses on vision and images. The human visual system has been widely studied from both behavioural and neuroscientific points of view, as vision is the dominant sense of most people. In turn, machine vision has also been an active area of research, currently dominated by the use of artificial neural networks. This work focuses on learning representations that are more aligned with…
Exploring Token-Level Augmentation in Vision Transformer for Semi-Supervised Semantic Segmentation
South China University of Technology（华南理工大学）；Pengcheng Laboratory（鹏城实验室）；China Tobacco Guangdong Industrial Corporation（广东中烟工业有限责任公司）；Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）
The COVID-19 pandemic has disrupted various levels of society. The use of masks is essential in preventing the spread of COVID-19 by identifying an image of a person using a mask. Although only 23.1% of people use masks correctly, Artificial Neural Networks (ANN) can help classify the use of good masks to help slow the spread of the Covid-19 virus. However, it requires a large dataset to train an ANN that can classify the use of masks correctly. MaskedFace-Net is a suitable dataset consisting of 137016 digital images with 4 class labels, namely Mask, Mask Chin, Mask Mouth Chin, and Mask Nose M…
Recently, the vision transformer (ViT) has made breakthroughs in image recognition. Its self-attention mechanism (MSA) can extract discriminative labeling information of different pixel blocks to improve image classification accuracy. However, the classification marks in their deep layers tend to ignore local features between layers. In addition, the embedding layer will be fixed-size pixel blocks. Input network Inevitably introduces additional image noise. To this end, we study a data augmentation vision transformer (DAVT) based on data augmentation and proposes a data augmentation method for…
Region modification-based data augmentation techniques have shown to improve performance for high level vision tasks (object detection, semantic segmentation, image classification, etc.) by encouraging underlying algorithms to focus on multiple discriminative features. However, as these techniques destroy spatial relationship with neighboring regions, performance can be deteriorated when using them to train algorithms designed for low level vision tasks (low light image enhancement, image dehazing, deblurring, etc.) where textural consistency between recovered and its neighboring regions is im…
Ford Greenfield Labs（福特格林菲尔德实验室）；University of Michigan（密歇根大学）
Real-world autonomous driving datasets comprise of images aggregated from different drives on the road. The ability to relight captured scenes to unseen lighting conditions, in a controllable manner, presents an opportunity to augment datasets with a richer variety of lighting conditions, similar to what would be encountered in the real-world. This paper presents a novel image-based relighting pipeline, SIMBAR, that can work with a single image as input. To the best of our knowledge, there is no prior work on scene relighting leveraging explicit geometric representations from a single image. W…
Achieving robustness to distributional shift is a longstanding and challenging goal of computer vision. Data augmentation is a commonly used approach for improving robustness, however robustness gains are typically not uniform across corruption types. Indeed increasing performance in the presence of random noise is often met with reduced performance on other corruptions such as contrast change. Understanding when and why these sorts of trade-offs occur is a crucial step towards mitigating them. Towards this end, we investigate recently observed trade-offs caused by Gaussian data augmentation a…
Vision Transformers (ViT) have been shown to attain highly competitive performance for a wide range of vision applications, such as image classification, object detection and semantic image segmentation. In comparison to convolutional neural networks, the Vision Transformer's weaker inductive bias is generally found to cause an increased reliance on model regularization or data augmentation ("AugReg" for short) when training on smaller training datasets. We conduct a systematic empirical study in order to better understand the interplay between the amount of training data, AugReg, model size a…
CNRS（法国国家科学研究中心）；Pascal institute - University Clermont-Ferrand（克莱蒙费朗大学帕斯卡研究所）；Frankfurt Institute for Advanced Studies（法兰克福高等研究所）
Biological vision systems are unparalleled in their ability to learn visual representations without supervision. In machine learning, self-supervised learning (SSL) has led to major advances in forming object representations in an unsupervised fashion. Such systems learn representations invariant to augmentation operations over images, like cropping or flipping. In contrast, biological vision systems exploit the temporal structure of the visual experience during natural interactions with objects. This gives access to "augmentations" not commonly used in SSL, like watching the same object from …
Indian Institute of Science（印度科学学院）；Google Research（谷歌研究院）
Recent work has shown that deep vision models tend to be overly dependent on low-level or "texture" features, leading to poor generalization. Various data augmentation strategies have been proposed to overcome this so-called texture bias in DNNs. We propose a simple, lightweight adversarial augmentation technique that explicitly incentivizes the network to learn holistic shapes for accurate prediction in an object classification setting. Our augmentations superpose edgemaps from one image onto another image with shuffled patches, using a randomly determined mixing proportion, with the image la…
