SkillOpt: Executive Strategy for Self-Evolving Agent Skills

Agent skills today are hand-crafted, generated one-shot, or evolved through loosely controlled self-revision, none of which behaves like a deep-learning optimizer for the skill, and none of which reliably improves over its starting point under feedback. We argue the skill should instead be trained as the external state of a frozen agent, with the same discipline that makes weight-space optimization reproducible. SkillOpt is, to our knowledge, the first systematic controllable text-space optimizer for agent skills: a separate optimizer model turns scored rollouts into bounded add/delete/replace edits on a single skill document, and an edit is accepted only when it strictly improves a held-out validation score. A textual learning-rate budget, rejected-edit buffer, and epoch-wise slow/meta update make skill training stable while adding zero inference-time model calls at deployment. Across six benchmarks, seven target models, and three execution harnesses (direct chat, Codex, Claude Code), SkillOpt is best or tied on all 52 evaluated (model, benchmark, harness) cells and beats every per-cell competitor among human, one-shot LLM, Trace2Skill, TextGrad, GEPA, and EvoSkill skills. On GPT-5.5 it lifts the average no-skill accuracy by +23.5 points in direct chat, by +24.8 inside the Codex agentic loop, and by +19.1 inside Claude Code. Transfer experiments further show that optimized skill artifacts retain value when moved across model scales, between Codex and Claude Code execution environments, and to a nearby math benchmark without further optimization. Code: https://aka.ms/skillopt

From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent Skills

Language agents increasingly improve by reusing \emph{skills} -- structured procedural artifacts distilled from past experience. In particular, \emph{domain-level} and \emph{model-generated} skills are especially promising. They offer fast adaptation within a domain by encoding domain-specific recurring procedures, and they scale beyond labor-intensive hand-crafting. However, while extraction methods continue to proliferate, understanding remains limited, with no comprehensive study spanning the full skill lifecycle -- \textbf{experience generation}, \textbf{skill extraction}, and \textbf{skill consumption} -- to ask whether such skills actually work, when they work, and what makes them succeed or fail. To close this gap, we build a utility-grounded evaluation framework that provides systematic experimental results across extractors and target agents, covering five diverse agentic task domains. We find that model-generated skills are beneficial on average but exhibit non-trivial negative transfer, and that neither extractors nor targets behave uniformly. A model can be a strong extractor yet a weak consumer, or vice versa, with skill utility independent of model scale or baseline task strength. To explain these patterns, we then dissect each lifecycle stage in depth, analyzing how experience composition shapes skill quality, what properties characterize useful skills, and how the same skill transfers across different consumers. Finally, we translate these findings into a concrete \emph{meta-skill} that guides skill extraction toward the features tied to actual utility, which consistently improves skill quality across domains and substantially reduces negative transfer.

Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models

We introduce Lens, a 3.8B-parameter T2I model that achieves performance competitive with, and in several cases surpassing, state-of-the-art models with more than 6B parameters across various benchmarks, while requiring significantly less training compute. For example, Lens requires only about 19.3% of the training compute used by Z-Image. The training efficiency of Lens stems from two key strategies beyond its compact model size. First, we maximize data information density per training batch by (i) training on Lens-800M, a dataset of 800M densely captioned image-text pairs whose captions are generated by GPT-4.1 and contain approximately 109 words on average, providing richer semantic supervision than conventional short captions, and (ii) constructing each batch from images with multiple resolutions and diverse aspect ratios, thereby enlarging the effective visual coverage of each optimization step. Second, we improve convergence speed through careful architectural choices, including adopting a semantic VAE that provides better latent representations and employing a strong language encoder that accelerates optimization while enabling multilingual generalization from English-only training data. After pre-training, we apply RL with taxonomy-driven prompts (Lens-RL-8K) and structured reward rubrics to suppress artifacts and improve visual quality, a reasoner module with training-free system prompt search to better align user requests with the model, and distillation-based acceleration for 4-step inference. Through efficient training and systematic optimization, Lens generalizes to arbitrary aspect ratios from 1:2 to 2:1 and resolutions up to 1440^2, and supports prompts in several commonly used languages. Thanks to its compact size, Lens generates a 1024^2 image in 3.15 seconds on a single NVIDIA H100 GPU, while its distilled turbo version performs 4-step generation in 0.84 seconds.

LACON: Training Text-to-Image Model from Uncurated Data

The success of modern text-to-image generation is largely attributed to massive, high-quality datasets. Currently, these datasets are curated through a filter-first paradigm that aggressively discards low-quality raw data based on the assumption that it is detrimental to model performance. Is the discarded bad data truly useless, or does it hold untapped potential? In this work, we critically re-examine this question. We propose LACON (Labeling-and-Conditioning), a novel training framework that exploits the underlying uncurated data distribution. Instead of filtering, LACON re-purposes quality signals, such as aesthetic scores and watermark probabilities as explicit, quantitative condition labels. The generative model is then trained to learn the full spectrum of data quality, from bad to good. By learning the explicit boundary between high- and low-quality content, LACON achieves superior generation quality compared to baselines trained only on filtered data using the same compute budget, proving the significant value of uncurated data.

Position: Spectral GNNs Are Neither Spectral Nor Superior for Node Classification

Spectral Graph Neural Networks (Spectral GNNs) for node classification promise frequency-domain filtering on graphs, yet rest on flawed foundations. Recent work shows that graph Laplacian eigenvectors do not in general have the key properties of a true Fourier basis, but leaves the empirical success of Spectral GNNs unexplained. We identify two theoretical glitches: (1) commonly used "graph Fourier bases" are not classical Fourier bases for graph signals; (2) (n-1)-degree polynomials (n = number of nodes) can exactly interpolate any spectral response via a Vandermonde system, so the usual "polynomial approximation" narrative is not theoretically justified. The effectiveness of GCN is commonly attributed to spectral low-pass filtering, yet we prove that low- and high-pass behaviors arise solely from message-passing dynamics rather than Graph Fourier Transform-based spectral formulations. We then analyze two representative directed spectral models, MagNet and HoloNet. Their reported effectiveness is not spectral: it arises from implementation issues that reduce them to powerful MPNNs. When implemented consistently with the claimed spectral algorithms, performance becomes weak. This position paper argues that: for node classification, Spectral GNNs neither meaningfully capture the graph spectrum nor reliably improve performance; competitive results are better explained by their equivalence to MPNNs, sometimes aided by implementations inconsistent with their intended design.

VISTA: Enhancing Visual Conditioning via Track-Following Preference Optimization in Vision-Language-Action Models

Vision-Language-Action (VLA) models have demonstrated strong performance across a wide range of robotic manipulation tasks. Despite the success, extending large pretrained Vision-Language Models (VLMs) to the action space can induce vision-action misalignment, where action predictions exhibit weak dependence on the current visual state, leading to unreliable action outputs. In this work, we study VLA models through the lens of visual conditioning and empirically show that successful rollouts consistently exhibit stronger visual dependence than failed ones. Motivated by this observation, we propose a training framework that explicitly strengthens visual conditioning in VLA models. Our approach first aligns action prediction with visual input via preference optimization on a track-following surrogate task, and then transfers the enhanced alignment to instruction-following task through latent-space distillation during supervised finetuning. Without introducing architectural modifications or additional data collection, our method improves both visual conditioning and task performance for discrete OpenVLA, and further yields consistent gains when extended to the continuous OpenVLA-OFT setting. Project website: https://vista-vla.github.io/ .

Bridging Online and Offline RL: Contextual Bandit Learning for Multi-Turn Code Generation

Recently, there have been significant research interests in training large language models (LLMs) with reinforcement learning (RL) on real-world tasks, such as multi-turn code generation. While online RL tends to perform better than offline RL, its higher training cost and instability hinders wide adoption. In this paper, we build on the observation that multi-turn code generation can be formulated as a one-step recoverable Markov decision process and propose contextual bandit learning with offline trajectories (Cobalt), a new method that combines the benefits of online and offline RL. Cobalt first collects code generation trajectories using a reference LLM and divides them into partial trajectories as contextual prompts. Then, during online bandit learning, the LLM is trained to complete each partial trajectory prompt through single-step code generation. Cobalt outperforms two multi-turn online RL baselines based on GRPO and VeRPO, and substantially improves R1-Distill 8B and Qwen3 8B by up to 9.0 and 6.2 absolute Pass@1 scores on LiveCodeBench. Also, we analyze LLMs' in-context reward hacking behaviors and augment Cobalt training with perturbed trajectories to mitigate this issue. Overall, our results demonstrate Cobalt as a promising solution for iterative decision-making tasks like multi-turn code generation. Our code and data are available at https://github.com/OSU-NLP-Group/cobalt.

SHARE: A Fully Unsupervised Framework for Single Hyperspectral Image Restoration

Hyperspectral image (HSI) restoration is a fundamental challenge in computational imaging and computer vision. It involves ill-posed inverse problems, such as inpainting and super-resolution. Although deep learning methods have transformed the field through data-driven learning, their effectiveness hinges on access to meticulously curated ground-truth datasets. This fundamentally restricts their applicability in real-world scenarios where such data is unavailable. This paper presents SHARE (Single Hyperspectral Image Restoration with Equivariance), a fully unsupervised framework that unifies geometric equivariance principles with low-rank spectral modelling to eliminate the need for ground truth. SHARE's core concept is to exploit the intrinsic invariance of hyperspectral structures under differentiable geometric transformations (e.g. rotations and scaling) to derive self-supervision signals through equivariance consistency constraints. Our novel Dynamic Adaptive Spectral Attention (DASA) module further enhances this paradigm shift by explicitly encoding the global low-rank property of HSI and adaptively refining local spectral-spatial correlations through learnable attention mechanisms. Extensive experiments on HSI inpainting and super-resolution tasks demonstrate the effectiveness of SHARE. Our method outperforms many state-of-the-art unsupervised approaches and achieves performance comparable to that of supervised methods. We hope that our approach will shed new light on HSI restoration and broader scientific imaging scenarios. The code will be released at https://github.com/xuwayyy/SHARE.

Equivariant Learning for Unsupervised Image Dehazing

Image Dehazing (ID) aims to produce a clear image from an observation contaminated by haze. Current ID methods typically rely on carefully crafted priors or extensive haze-free ground truth, both of which are expensive or impractical to acquire, particularly in the context of scientific imaging. We propose a new unsupervised learning framework called Equivariant Image Dehazing (EID) that exploits the symmetry of image signals to restore clarity to hazy observations. By enforcing haze consistency and systematic equivariance, EID can recover clear patterns directly from raw, hazy images. Additionally, we propose an adversarial learning strategy to model unknown haze physics and facilitate EID learning. Experiments on two scientific image dehazing benchmarks (including cell microscopy and medical endoscopy) and on natural image dehazing have demonstrated that EID significantly outperforms state-of-the-art approaches. By unifying equivariant learning with modelling haze physics, we hope that EID will enable more versatile and effective haze removal in scientific imaging. Code and datasets will be published.

ReasonGen-R1: CoT for Autoregressive Image generation models through SFT and RL

Although chain-of-thought reasoning and reinforcement learning (RL) have driven breakthroughs in NLP, their integration into generative vision models remains underexplored. We introduce ReasonGen-R1, a two-stage framework that first imbues an autoregressive image generator with explicit text-based "thinking" skills via supervised fine-tuning on a newly generated reasoning dataset of written rationales, and then refines its outputs using Group Relative Policy Optimization. To enable the model to reason through text before generating images, We automatically generate and release a corpus of model crafted rationales paired with visual prompts, enabling controlled planning of object layouts, styles, and scene compositions. Our GRPO algorithm uses reward signals from a pretrained vision language model to assess overall visual quality, optimizing the policy in each update. Evaluations on GenEval, DPG, and the T2I benchmark demonstrate that ReasonGen-R1 consistently outperforms strong baselines and prior state-of-the-art models. More: aka.ms/reasongen.