arXiv:2603.07131v1 Announce Type: cross Abstract: Large Vision Language Models (LVLMs) show immense potential for automated ophthalmic diagnosis. However, their clinical deployment is severely hindered by lacking domain-specific knowledge. In this work, we identify two structural deficiencies hindering reliable medical reasoning: 1) the Perception Gap, where general-purpose visual encoders fail to resolve fine-grained pathological cues (e.g., microaneurysms); and 2) the Reasoning Gap, where sparse visual evidence is progressively overridden by massive language priors in deeper transformer layers, leading to ungrounded hallucinations. To bridge these gaps, we propose EyExIn, a data-efficient framework designed to anchor retinal VLMs with expert knowledge via a Deep Expert Injection mechanism. Our architecture employs an Expert-Aware Dual-Stream encoding strategy that decouples visual representation into a general stream for anatomical context and a specialized expert stream for pathological semantics. To ensure high-fidelity integration, we design a Semantic-Adaptive Gated Fusion module, which dynamically amplifies subtle lesion signals while filtering irrelevant background noise. Furthermore, we introduce Adaptive Deep Expert Injection to embed persistent "Vision Anchors" by integrating fused visual features as residual biases directly into intermediate LLM layers. This mechanism creates a visual shortcut that forces the reasoning stack to remain strictly grounded in visual evidence. Extensive experiments across four benchmarks demonstrate that our model consistently outperforms massive proprietary systems. EyExIn significantly enhances domain-specific knowledge embedding and achieves state-of-the-art precision in ophthalmic visual question answering, advancing the development of trustworthy ophthalmic AI.

arXiv:2603.03818v1 Announce Type: cross Abstract: Continual learning is a long-standing challenge in robot policy learning, where a policy must acquire new skills over time without catastrophically forgetting previously learned ones. While prior work has extensively studied continual learning in relatively small behavior cloning (BC) policy models trained from scratch, its behavior in modern large-scale pretrained Vision-Language-Action (VLA) models remains underexplored. In this work, we found that pretrained VLAs are remarkably resistant to forgetting compared with smaller policy models trained from scratch. Simple Experience Replay (ER) works surprisingly well on VLAs, sometimes achieving zero forgetting even with a small replay data size. Our analysis reveals that pretraining plays a critical role in downstream continual learning performance: large pretrained models mitigate forgetting with a small replay buffer size while maintaining strong forward learning capabilities. Furthermore, we found that VLAs can retain relevant knowledge from prior tasks despite performance degradation during learning new tasks. This knowledge retention enables rapid recovery of seemingly forgotten skills through finetuning. Together, these insights imply that large-scale pretraining fundamentally changes the dynamics of continual learning, enabling models to continually acquire new skills over time with simple replay. Code and more information can be found at https://ut-austin-rpl.github.io/continual-vla

arXiv:2603.01620v2 Announce Type: replace Abstract: Tool-integrated agents that interleave reasoning with API calls are promising for complex tasks, yet aligning them for high-stakes, domain-specific deployment remains challenging: existing reinforcement learning approaches rely on coarse binary rewards that cannot distinguish tool selection errors from malformed parameters. We present ToolRLA, a three-stage post-training pipeline (SFT $\rightarrow$ GRPO $\rightarrow$ DPO) for domain-specific tool agents. The core contribution is a fine-grained reward function with multiplicative correctness decomposition spanning four dimensions -- format validity, tool selection, parameter accuracy, and regulatory compliance -- that encodes domain priority orderings as inductive biases in the reward landscape. Deployed on a financial advisory copilot (80+ advisors, 1,200+ daily queries), ToolRLA achieves over three months: a 47\% improvement in task completion rate ($62\%\rightarrow91\%$), a 63\% reduction in tool invocation errors ($38\%\rightarrow14\%$), and a 93\% reduction in regulatory violations ($12\%\rightarrow0.8\%$), within sub-2-second latency. Ablation studies show the multiplicative reward design accounts for 7 percentage points of improvement over additive alternatives. Generalization is further validated on ToolBench and API-Bank.

arXiv:2509.25541v2 Announce Type: replace-cross Abstract: Although reinforcement learning (RL) has emerged as a promising approach for improving vision-language models (VLMs) and multimodal large language models (MLLMs), current methods rely heavily on manually curated datasets and costly human verification, which limits scalable self-improvement in multimodal systems. To address this challenge, we propose Vision-Zero, a label-free, domain-agnostic multi-agent self-play framework for self-evolving VLMs through competitive visual games generated from arbitrary image inputs. Specifically, Vision-Zero encompasses three main attributes: (1) Strategic Self-Play Framework: Vision-Zero trains VLMs in "Who Is the Spy"-style games, where the models engage in strategic reasoning and actions across multiple roles. Through interactive gameplay, models autonomously generate their training data without human annotation. (2) Gameplay from Arbitrary Images: Unlike existing gamified frameworks, Vision-Zero can generate games from arbitrary images, thereby enhancing the model's reasoning ability across diverse domains and showing strong generalization to different tasks. We demonstrate this versatility using three distinct types of image datasets: CLEVR-based synthetic scenes, charts, and real-world images. (3) Sustainable Performance Gain: We introduce Iterative Self-Play Policy Optimization (Iterative-SPO), a novel training algorithm that alternates between Self-Play and reinforcement learning with verifiable rewards (RLVR), mitigating the performance plateau often seen in self-play-only training and achieving sustained long-term improvements. Despite using label-free data, Vision-Zero achieves state-of-the-art performance on reasoning, chart question answering, and vision-centric understanding tasks, surpassing other annotation-based methods. Models and code have been released at https://github.com/wangqinsi1/Vision-Zero.

arXiv:2410.13331v2 Announce Type: replace-cross Abstract: The Straight-Through Estimator (STE) is the dominant method for training neural networks with discrete variables, enabling gradient-based optimisation by routing gradients through a differentiable surrogate. However, existing STE variants conflate two fundamentally distinct concerns: forward-pass stochasticity, which controls exploration and latent space utilisation, and backward-pass gradient dispersion i.e how learning signals are distributed across categories. We show that these concerns are qualitatively different and that tying them to a single temperature parameter leaves significant performance gains untapped. We propose Decoupled Straight-Through (Decoupled ST), a minimal modification that introduces separate temperatures for the forward pass ($\tau_f$) and the backward pass ($\tau_b$). This simple change enables independent tuning of exploration and gradient dispersion. Across three diverse tasks (Stochastic Binary Networks, Categorical Autoencoders, and Differentiable Logic Gate Networks), Decoupled ST consistently outperforms Identity STE, Softmax STE, and Straight-Through Gumbel-Softmax. Crucially, optimal $(\tau_f, \tau_b)$ configurations lie far off the diagonal $\tau_f = \tau_b$, confirming that the two concerns do require different answers and that single-temperature methods are fundamentally constrained.

arXiv:2602.10604v2 Announce Type: replace-cross Abstract: We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

arXiv:2411.16085v4 Announce Type: replace-cross Abstract: AdamW has been the default optimizer for transformer pretraining. For many years, our community searched for faster and more stable optimizers with only constrained positive outcomes. In this work, we propose a \textbf{one-line modification in Pytorch} to any momentum-based optimizer, which we rename cautious optimizer, e.g. C-AdamW and C-Lion. Our theoretical result shows that this modification preserves Adam's Hamiltonian function and it does not break the convergence guarantee under the Lyapunov analysis. In addition, a whole new family of optimizers is revealed by our theoretical insight. Among them, we pick the simplest one for empirical experiments, showing not only consistent speed-up on LLM pretraining, but also image classification, with minimum extra tuning on hyperparameters. Code is available at https://github.com/kyleliang919/C-Optim.