Searcharxiv⌕ Search

arXiv subjects

Suozhi Huang

Publications and source records attributed to Suozhi Huang.

15 recordsLinked to original sources

Can Language Models Learn to Forecast Stock Prices?

We study whether language models can learn to forecast stock returns through training. We build BETA, a chronological sandbox over S&P 500 stocks in which a model calls tools to retrieve prices, indicators, and news available up to a decision date, and then forecasts the stock's future return. Evaluating frontier models in this sandbox reveals three weaknesses: their scores decline as the test period approaches the present while a simple momentum baseline stays stable, they predict gains far more often than gains occur, and they predict moves much smaller than those that are realized. We first fine-tune on filtered, hindsight-generated tool-use trajectories for historical tasks judged predictable from the evidence available at the time. We then apply reinforcement learning with rewards based on realized returns. During RL, long investigations frequently exceed the context window. We therefore introduce a selective compaction tool that allows the model to compress chosen portions of its interaction history. Selective compaction changes the context available to subsequent decisions, so policy updates must preserve the context in which each action was generated. Meanwhile, rollouts vary substantially in the number and length of their trainable turns, making a shared group-level baseline less effective. We address these challenges with Context-Aligned Turn-Level PPO, which uses a turn-level critic and computes policy updates using each turn's actual context. The resulting models, which we call AURA, substantially reduce all three weaknesses. Training improves the 8B model's score from 24.0 to 47.0 and the 30B model's from 27.4 to 54.7, with the latter surpassing all evaluated frontier models on our benchmark and achieving its largest margin on the most recent test period. These results suggest that agentic RL can improve performance on tasks with noisy, delayed real-world outcomes.

cs.CL↗

LoopFill: Pipelined Multi-Token Speculative Decoding for Looped Transformers

Looped Transformers improve parameter efficiency by repeatedly applying the same Transformer layers. During autoregressive decoding, however, generating each token requires several sequential passes through these layers, leading to high decoding latency. Pipelined self-speculative methods use shallow-loop predictions to process tokens at different loops in the same batch, but each token must reach the drafting depth before the next can enter the pipeline. We introduce LoopFill, a training-free speculative decoding algorithm for looped LLM that combines multi-token suffix drafts with shallow-loop predictions in a single draft tree. Both drafters contribute to one tree, so a suffix lookup can immediately extend a shallow prediction and several future tokens can enter the pipeline together. The shared block computes the highest-scoring draft candidates at different loop depths in one batch, and full-depth verification preserves the model's autoregressive output distribution. Across seven benchmarks, LoopFill's average speedup over greedy autoregressive decoding is $5.85\times$ on Ouro-2.6B and $5.13\times$ on Huginn-3.5B. The speedups exceed those of LoopAhead, the strongest baseline, by \(40\%\) on Ouro-2.6B and \(25\%\) on Huginn-3.5B.

cs.LG↗

VideoGen-Agent: Reinforcing Video Generation Agents

Recent advances in video generative models have enabled high-fidelity, temporally coherent video generation. However, these models often struggle to satisfy prompts requiring specialized knowledge, specific identities, physical consistency, or ordered events. In this paper, we present VideoGen-Agent, a multimodal agent trained through multitask agentic reinforcement learning to use external tools for video generation. The agent coordinates augmentation, generation, and verification tools through multi-turn interactions, using the prompt and intermediate observations to guide its decisions. We train a shared policy on a category-balanced dataset spanning six tasks. Supervised fine-tuning on teacher-generated trajectories establishes tool-use behavior, which is then refined through reinforcement learning. A category-aware hybrid reward evaluates tool-call validity, task-appropriate tool use, and generated video quality. We further introduce VABench, a held-out benchmark of 600 prompts covering procedural knowledge, single- and multi-entity identity preservation, physical consistency, scene composition, and multi-shot temporal structure. On VABench, VideoGen-Agent improves over its base text-to-video generator by 19.1 points, from 56.5 to 75.6. Upgrading the generation tools further raises the score to 86.1 without additional agent training. Human raters prefer the upgraded configuration over the strongest standalone baseline in 84.3% of comparisons. These results support learning tool use across video-generation tasks and show that the trained agent can benefit from subsequent advances in generation tools. Project page: https://andyca111.github.io/VideoGen_Agent/

cs.CV↗

LabWorld: Learning Interactive 4D Worlds for Scientific Experimentation

Autonomous scientific agents must not only reason about experiments but also interact with the physical environments in which they are executed. Existing laboratory simulators typically rely on manually constructed assets and predefined procedures, while laboratory videos capture real equipment, layouts, and interactions but leave this information implicit in pixels. We introduce LabWorld, an agent-driven Real2Sim framework for reconstructing interactive 4D laboratory worlds from monocular RGB video. LabWorld recovers reusable 3D assets, a metric scene, time-varying object states, and procedure events. Multi-frame observations provide complementary evidence for asset and scene reconstruction, while motion reconstruction combines visual and geometric observations with event constraints encoding the spatial and temporal requirements of laboratory interactions. We evaluate LabWorld on real pipetting and vortexing videos collected in our laboratory against general-purpose agentic and dedicated reconstruction baselines. For pipetting, LabWorld achieves a mean Mask IoU of 0.501, exceeding the strongest baseline by 0.033. For vortexing, it achieves a mean Tube-Length-Normalized PCK-AUC of 0.638, exceeding the strongest baseline by 0.035. On five pipetting videos, event-conditioned reconstruction increases the fraction of modeled physical constraints satisfied from 14.3% (5/35) to 100% (35/35). The resulting worlds expose geometry, motion, articulation, and events as explicit, editable state for replay, inspection, and future embodied scientific systems.

cs.CV↗

GraphCast-Mamba: Learning Non-Markovian Corrections for Weather Forecasting

Unresolved atmospheric processes can introduce history dependence beyond the short input windows used by many weather forecasting models. We introduce GraphCast--Mamba, which augments a frozen, pretrained $1^\circ$ GraphCast backbone with a trainable Mamba correction branch combining graph message passing and Mamba layers. The branch retains a temporal memory state across successive six-hour forecast steps, learning corrections informed by temporal context beyond GraphCast's two-frame input. Training is confined to the correction branch, using ERA5 data from 2015--2021. After 2,500 optimizer updates, including warm-up, the selected model achieves substantial improvements relative to the uncorrected GraphCast backbone. It reduces the GraphCast loss (normalized weighted mean-squared error; MSE) by 14.42% when aggregated over the full ten-day forecast and by 21.66% at day 10 (240 h after forecast initialization). At this final forecast lead time, root-mean-square error (RMSE) decreases by 7.09--14.60% across all 11 evaluated variable groups, spanning temperature, wind, pressure, geopotential, specific humidity, and precipitation. These results demonstrate broad improvements in ten-day weather forecasts through a short correction-training schedule while retaining the pretrained backbone.

cs.LG↗

JUICE: Unified Multimodal Modeling in Pretrained Representation Spaces

Most unified multimodal models treat images and text asymmetrically: a pretrained language model generates text, an attached diffusion head generates images, and outputs that contain both are produced in a fixed order, so the first output cannot adapt to the second. We ask whether both modalities can instead be generated by one process that is trained on them jointly from scratch. Pretrained encoder spaces hold both modalities in one continuous form and carry knowledge learned at scale, and recent work finds them effective as generative targets for vision and language. We therefore model the joint distribution directly in these spaces. We introduce JUICE (Joint Unified multimodal modeling In Continuous Encoder spaces), a single flow matching Transformer over frozen image and text representations, in which each modality has its own noise level. Which modalities are clean and which are generated then decides the task: image generation, captioning, visual question answering, or joint generation. Controlled pretraining runs test each design choice. At equal guidance, generating text as continuous states leaves image generation intact (48.41 GenEval, against 47.86 for a model trained only for text-to-image generation), while captions lose some exact wording. Training on joint and conditional denoising together beats either alone, and semantic encoder spaces beat VAE latents and raw pixels as targets. At 2.1B parameters, with no language model initialization and at 224x224, JUICE reaches up to 0.88 on GenEval and 73.1 on VQAv2. By sampling both modalities from the joint distribution (coupled sampling), it can also draw a route on a map while writing out the matching moves. On mazes with several valid solutions, joint success rises from 54.6% with independent sampling to 93.7%, and no modality has to go first. On visual spatial planning, JUICE solves 99.8% of maps at its training sizes.

cs.CV↗

From Gradients to Capabilities: Understanding Multi-Teacher On-Policy Distillation

Multi-teacher on-policy distillation (MOPD) aims to combine the strengths of RL-trained teachers in a single student, but how teacher signals affect parameter changes remains underexplored. We study Qwen3-1.7B with four domain teachers trained with RL from the same initialization as the student, comparing gradients, optimizer updates, and task learning curves, with additional SmolLM3-3B diagnostics. We find that several factors influence teacher signals. First, loss averaging implicitly weights responses: token averaging favors longer responses, and equalizing domain contributions retains this weighting within domains. Second, Adam's first moment reduces differences in parameter updates: the cosine similarity is 0.83 between teachers and 0.96 between averaging rules, despite differences in raw gradients. Third, BF16 rounding hides small changes: about 97\% of FP32 master weights differ from initialization, but only 7--11\% of BF16 weights do. Finally, the top-64 intersection KL gradient closely matches Qwen's full-vocabulary gradient, but the effect on task performance depends on averaging: mathematics accuracy is 2.6 points higher than with sampled-token policy-gradient (PG) under response averaging and 2.1 points lower under global token averaging.

cs.LG↗

AQuA: Recursively Self-Improving Quantitative Trading Research Agents

We study recursive self-improvement at the level of quantitative-investment research: whether an autonomous system can use evidence from earlier experiments to improve the hypotheses and candidates proposed in later iterations. We present AQuA, which comprises two separate language-model-driven research systems: one for symbolic factor discovery and one for trainable model development. Each system records experimental results and uses them to guide subsequent proposals. Each operates in a fixed sandbox, which fixes the data splits, feature and label definitions, and evaluator while allowing the model to act only through constrained factor expressions or configuration diffs. The factor system, a manager-mediated multi-agent pipeline, discovers and combines factors into a signal that reaches a combined validation information coefficient of about $0.190$ on a crypto universe. The model system, a config-driven loop over a hybrid time-series architecture, reaches a per-stock information coefficient of $+0.0843$ on US equities and converts it into a threshold long/short strategy with a held-out Sharpe of up to $+2.50$ at a two-leg cost. The strategy is positive in every year from 2021 to 2025.

cs.CL↗

Any2Poster: Any-Source Poster Generation Across Modalities and Domains

Visual posters are a compact medium for communicating dense information, yet progress on automatic poster generation remains difficult to measure because existing evaluations are often restricted to paper-only inputs, narrow domains, or surface-level visual similarity. We introduce Any2Poster Bench, a benchmark for any-source poster generation that evaluates systems across eight input modalities--PDFs, URLs, PPTX, DOCX, Markdown, LaTeX, notebooks, and videos--and five content domains. Any2Poster Bench pairs each source with quiz-based probes of verbatim factual retention and interpretive understanding, together with VLM-based judgments of visual quality, layout, readability, content completeness, and logical flow, enabling reproducible assessment of both information fidelity and visual communication. To instantiate and validate this benchmark, we further present Any2Poster Agent, an end-to-end reference agent that parses heterogeneous sources, organizes salient content, plans poster layouts, renders posters, and iteratively refines them using visual feedback. On Any2Poster Bench, Any2Poster Agent achieves 87.25% average accuracy across input modalities and 87.28% across content domains. On PaperQuiz-style evaluation, where prior paper-to-poster agents are directly comparable, Any2Poster Agent improves over PosterAgent-4o from 51.06-51.33% to 72.58% overall accuracy and from 116-121 to 145.16 in density-augmented score. Together, Any2Poster Bench and Any2Poster Agent provide a reusable evaluation resource and a competitive baseline for studying multimodal, domain-general poster generation.

cs.CV↗

CryptoBench: A Dynamic Benchmark for Expert-Level Evaluation of LLM Agents in Cryptocurrency

This paper introduces CryptoBench, the first expert-curated, dynamic benchmark designed to rigorously evaluate the real-world capabilities of Large Language Model (LLM) agents in the uniquely demanding and fast-paced cryptocurrency domain. Unlike general-purpose agent benchmarks for search and prediction, professional crypto analysis presents specific challenges: \emph{extreme time-sensitivity}, \emph{a highly adversarial information environment}, and the critical need to synthesize data from \emph{diverse, specialized sources}, such as on-chain intelligence platforms and real-time Decentralized Finance (DeFi) dashboards. CryptoBench thus serves as a much more challenging and valuable scenario for LLM agent assessment. To address these challenges, we constructed a live, dynamic benchmark featuring 50 questions per month, expertly designed by crypto-native professionals to mirror actual analyst workflows. These tasks are rigorously categorized within a four-quadrant system: Simple Retrieval, Complex Retrieval, Simple Prediction, and Complex Prediction. This granular categorization enables a precise assessment of an LLM agent's foundational data-gathering capabilities alongside its advanced analytical and forecasting skills. Our evaluation of ten LLMs, both directly and within an agentic framework, reveals a performance hierarchy and uncovers a failure mode. We observe a \textit{retrieval-prediction imbalance}, where many leading models, despite being proficient at data retrieval, demonstrate a pronounced weakness in tasks requiring predictive analysis. This highlights a problematic tendency for agents to appear factually grounded while lacking the deeper analytical capabilities to synthesize information.

cs.CL↗

LeanProgress: Guiding Search for Neural Theorem Proving via Proof Progress Prediction

Mathematical reasoning remains a significant challenge for Large Language Models (LLMs) due to hallucinations. When combined with formal proof assistants like Lean, these hallucinations can be eliminated through rigorous verification, making theorem proving reliable. However, even with formal verification, LLMs still struggle with long proofs and complex mathematical formalizations. While Lean with LLMs offers valuable assistance with retrieving lemmas, generating tactics, or even complete proofs, it lacks a crucial capability: providing a sense of proof progress. This limitation particularly impacts the overall development efficiency in large formalization projects. We introduce LeanProgress, a method that predicts the progress in the proof. Training and evaluating our models made on a large corpus of Lean proofs from Lean Workbook Plus and Mathlib4 and how many steps remain to complete it, we employ data preprocessing and balancing techniques to handle the skewed distribution of proof lengths. Our experiments show that LeanProgress achieves an overall prediction accuracy of 75.8% in predicting the amount of progress and, hence, the remaining number of steps. When integrated into a best-first search framework using Reprover, our method shows a 3.8% improvement on Mathlib4 compared to baseline performances of 41.4%, particularly for longer proofs. These results demonstrate how proof progress prediction can enhance both automated and interactive theorem proving, enabling users to make more informed decisions about proof strategies. Our code is merged in this library here https://github.com/lean-dojo/LeanDojo-v2.

cs.AI↗

FutureX-Pro: Extending Future Prediction to High-Value Vertical Domains

Building upon FutureX, which established a live benchmark for general-purpose future prediction, this report introduces FutureX-Pro, including FutureX-Finance, FutureX-Retail, FutureX-PublicHealth, FutureX-NaturalDisaster, and FutureX-Search. These together form a specialized framework extending agentic future prediction to high-value vertical domains. While generalist agents demonstrate proficiency in open-domain search, their reliability in capital-intensive and safety-critical sectors remains under-explored. FutureX-Pro targets four economically and socially pivotal verticals: Finance, Retail, Public Health, and Natural Disaster. We benchmark agentic Large Language Models (LLMs) on entry-level yet foundational prediction tasks -- ranging from forecasting market indicators and supply chain demands to tracking epidemic trends and natural disasters. By adapting the contamination-free, live-evaluation pipeline of FutureX, we assess whether current State-of-the-Art (SOTA) agentic LLMs possess the domain grounding necessary for industrial deployment. Our findings reveal the performance gap between generalist reasoning and the precision required for high-value vertical applications.

cs.AI↗

InternLM2.5-StepProver: Advancing Automated Theorem Proving via Critic-Guided Search

Large Language Models (LLMs) have emerged as powerful tools in mathematical theorem proving, particularly when utilizing formal languages such as LEAN. A prevalent proof method involves the LLM prover iteratively constructing the proof tactic by tactic, typically following a best-first search scheme. However, this method often ignores the critical preference information inside the existing tactic trajectories, hindering the search for deeper proofs. We propose an intuitive yet effective method, which utilizes a critic model to capture the preference information and to guide the search of the prover model at runtime. Given the prover-critic framework, a large-scale expert iteration with more than 20,000 CPU days is then applied to further fine-tune the prover and the critic. The trained InternLM2.5-StepProver critic significantly boosts the performance of the prover model (59.4% to 65.9%). We also analyze the impact of the critic on various aspects of the theorem proving process during expert iteration, providing insights into its effectiveness. We open-source our models and searched proofs at https://github.com/InternLM/InternLM-Math and https://huggingface.co/datasets/internlm/Lean-Workbook.

cs.AI↗

Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free

Gating mechanisms have been widely utilized, from early models like LSTMs and Highway Networks to recent state space models, linear attention, and also softmax attention. Yet, existing literature rarely examines the specific effects of gating. In this work, we conduct comprehensive experiments to systematically investigate gating-augmented softmax attention variants. Specifically, we perform a comprehensive comparison over 30 variants of 15B Mixture-of-Experts (MoE) models and 1.7B dense models trained on a 3.5 trillion token dataset. Our central finding is that a simple modification-applying a head-specific sigmoid gate after the Scaled Dot-Product Attention (SDPA)-consistently improves performance. This modification also enhances training stability, tolerates larger learning rates, and improves scaling properties. By comparing various gating positions and computational variants, we attribute this effectiveness to two key factors: (1) introducing non-linearity upon the low-rank mapping in the softmax attention, and (2) applying query-dependent sparse gating scores to modulate the SDPA output. Notably, we find this sparse gating mechanism mitigates 'attention sink' and enhances long-context extrapolation performance, and we also release related $\href{https://github.com/qiuzh20/gated_attention}{codes}$ and $\href{https://huggingface.co/QwQZh/gated_attention}{models}$ to facilitate future research.

cs.CL↗

ActFormer: Scalable Collaborative Perception via Active Queries

Collaborative perception leverages rich visual observations from multiple robots to extend a single robot's perception ability beyond its field of view. Many prior works receive messages broadcast from all collaborators, leading to a scalability challenge when dealing with a large number of robots and sensors. In this work, we aim to address \textit{scalable camera-based collaborative perception} with a Transformer-based architecture. Our key idea is to enable a single robot to intelligently discern the relevance of the collaborators and their associated cameras according to a learned spatial prior. This proactive understanding of the visual features' relevance does not require the transmission of the features themselves, enhancing both communication and computation efficiency. Specifically, we present ActFormer, a Transformer that learns bird's eye view (BEV) representations by using predefined BEV queries to interact with multi-robot multi-camera inputs. Each BEV query can actively select relevant cameras for information aggregation based on pose information, instead of interacting with all cameras indiscriminately. Experiments on the V2X-Sim dataset demonstrate that ActFormer improves the detection performance from 29.89% to 45.15% in terms of AP@0.7 with about 50% fewer queries, showcasing the effectiveness of ActFormer in multi-agent collaborative 3D object detection.

cs.CV↗