arXiv subjects

Siyu Xia

Publications and source records attributed to Siyu Xia.

At least 19 recordsLinked to original sources

arXiv2026

UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task Streams

Memory is essential for LLM agents to accumulate task experience and reuse task-specific execution strategies. However, real-world deployment over boundary-agnostic and evolving task streams exposes a fundamental stability-plasticity dilemma. External retrieval-based memory can rapidly absorb new evidence, but it often fails to internalize recurring execution patterns and incurs inference-time retrieval overhead. Parametric memory enables stable and efficient execution once learned, but typically relies on explicit task boundaries and fixed parameter budgets. Inspired by the human brain, which balances plasticity and stability through complementary episodic storage and gradual consolidation, we propose UniMem, a self-routing framework for autonomous memory management. UniMem uses learnable routing tokens as memory controllers, enabling adaptive coordination between complementary memory pathways: novel or sparse tasks are retained in an episodic buffer for retrieval-augmented execution, while recurring and reliable patterns are consolidated into expandable parametric memory. By decoupling task identification from task execution with routing tokens and parametric memory blocks, UniMem expands memory on demand without task labels during deployment or uncontrolled parameter growth. Experiments on long-horizon streaming task sequences show that UniMem consistently outperforms baselines while maintaining execution fidelity, achieving an average gain of 4.0 EM points across three backbone models.

cs.CL↗

arXiv2026

PSHuman: Photorealistic Single-image 3D Human Reconstruction using Cross-Scale Multiview Diffusion and Explicit Remeshing

Detailed and photorealistic 3D human modeling is essential for various applications and has seen tremendous progress. However, full-body reconstruction from a monocular RGB image remains challenging due to the ill-posed nature of the problem and sophisticated clothing topology with self-occlusions. In this paper, we propose PSHuman, a novel framework that explicitly reconstructs human meshes utilizing priors from the multiview diffusion model. It is found that directly applying multiview diffusion on single-view human images leads to severe geometric distortions, especially on generated faces. To address it, we propose a cross-scale diffusion that models the joint probability distribution of global full-body shape and local facial characteristics, enabling detailed and identity-preserved novel-view generation without any geometric distortion. Moreover, to enhance cross-view body shape consistency of varied human poses, we condition the generative model on parametric models like SMPL-X, which provide body priors and prevent unnatural views inconsistent with human anatomy. Leveraging the generated multi-view normal and color images, we present SMPLX-initialized explicit human carving to recover realistic textured human meshes efficiently. Extensive experimental results and quantitative evaluations on CAPE and THuman2.1 datasets demonstrate PSHumans superiority in geometry details, texture fidelity, and generalization capability.

cs.CV↗

arXiv2025

From Experience to Strategy: Empowering LLM Agents with Trainable Graph Memory

Large Language Models (LLMs) based agents have demonstrated remarkable potential in autonomous task-solving across complex, open-ended environments. A promising approach for improving the reasoning capabilities of LLM agents is to better utilize prior experiences in guiding current decisions. However, LLMs acquire experience either through implicit memory via training, which suffers from catastrophic forgetting and limited interpretability, or explicit memory via prompting, which lacks adaptability. In this paper, we introduce a novel agent-centric, trainable, multi-layered graph memory framework and evaluate how context memory enhances the ability of LLMs to utilize parametric information. The graph abstracts raw agent trajectories into structured decision paths in a state machine and further distills them into high-level, human-interpretable strategic meta-cognition. In order to make memory adaptable, we propose a reinforcement-based weight optimization procedure that estimates the empirical utility of each meta-cognition based on reward feedback from downstream tasks. These optimized strategies are then dynamically integrated into the LLM agent's training loop through meta-cognitive prompting. Empirically, the learnable graph memory delivers robust generalization, improves LLM agents' strategic reasoning performance, and provides consistent benefits during Reinforcement Learning (RL) training.

cs.CL↗

arXiv2025

MTIR-SQL: Multi-turn Tool-Integrated Reasoning Reinforcement Learning for Text-to-SQL

As large language models (LLMs) are increasingly used in Text-to-SQL tasks, Reinforcement Learning (RL) has become a common method for improving performance. Existing methods primarily rely on static execution feedback, which restricts real-time error correction. However, integrating multi-turn tool invocation along with dynamic feedback could significantly improve adaptability and robustness, ultimately enhancing model performance. To address these issues, we propose MTIR-SQL, an innovative Multi-turn Tool-Integrated Reasoning reinforcement learning framework for Text-to-SQL. Our approach introduces an execution-aware multi-turn reasoning paradigm that seamlessly incorporates database execution feedback at each reasoning step, enabling context-sensitive query generation and progressive refinement throughout the reasoning process. The framework extends the GRPO algorithm to accommodate complex multi-turn interaction scenarios. Considering the training instability characteristics of MTIR and the potential for significant Deviation of model distribution from the initial model, we enhance the GRPO algorithm by adding a trajectory filtering mechanism and removing KL loss constraints. Experimental results demonstrate that MTIR-SQL, with 4B parameters, achieves \textbf{64.4}\% accuracy in the BIRD Dev and 84.6% execution accuracy in the SPIDER Dev, significantly outperforming existing approaches.

cs.AI↗

arXiv2025

Internal Chain-of-Thought: Empirical Evidence for Layer-wise Subtask Scheduling in LLMs

We show that large language models (LLMs) exhibit an $\textit{internal chain-of-thought}$: they sequentially decompose and execute composite tasks layer-by-layer. Two claims ground our study: (i) distinct subtasks are learned at different network depths, and (ii) these subtasks are executed sequentially across layers. On a benchmark of 15 two-step composite tasks, we employ layer-from context-masking and propose a novel cross-task patching method, confirming (i). To examine claim (ii), we apply LogitLens to decode hidden states, revealing a consistent layerwise execution pattern. We further replicate our analysis on the real-world $\text{TRACE}$ benchmark, observing the same stepwise dynamics. Together, our results enhance LLMs transparency by showing their capacity to internally plan and execute subtasks (or instructions), opening avenues for fine-grained, instruction-level activation steering.

cs.CL↗

arXiv2025

RLFactory: A Plug-and-Play Reinforcement Learning Post-Training Framework for LLM Multi-Turn Tool-Use

Large language models excel at basic reasoning but struggle with tasks that require interaction with external tools. We present RLFactory, a plug-and-play reinforcement learning post-training framework for multi-round tool use. RLFactory tackles (i) tool-call stability and adaptability amid tool heterogeneity and interface issues via an asyncio-based asynchronous caller and a decoupled tool/training architecture, and (ii) diverse evaluation needs via a reward layer supporting rule-based, model-judgment, and tool-verification signals. It reconstructs the MDP by introducing observation markers from tool feedback, closing the loop among model, tools, and environment, and implements a generate-parse-invoke-update workflow for dynamic policy optimization. On Search-R1 with Qwen3-4B, RLFactory achieves a 0.486 test score on the Natural Questions (NQ) dataset, surpassing larger models trained with similar techniques (e.g., Qwen2.5-7B-Instruct-GRPO at 0.473), and increases training throughput by 6.8x. RLFactory provides a low-barrier, highly adaptable framework for strengthening multi-round tool use of LLMs in real-world scenarios. Code: https://github.com/Simple-Efficient/RL-Factory.

cs.LG↗

arXiv2025

EconGym: A Scalable AI Testbed with Diverse Economic Tasks

Artificial intelligence (AI) has become a powerful tool for economic research, enabling large-scale simulation and policy optimization. However, applying AI effectively requires simulation platforms for scalable training and evaluation-yet existing environments remain limited to simplified, narrowly scoped tasks, falling short of capturing complex economic challenges such as demographic shifts, multi-government coordination, and large-scale agent interactions. To address this gap, we introduce EconGym, a scalable and modular testbed that connects diverse economic tasks with AI algorithms. Grounded in rigorous economic modeling, EconGym implements 11 heterogeneous role types (e.g., households, firms, banks, governments), their interaction mechanisms, and agent models with well-defined observations, actions, and rewards. Users can flexibly compose economic roles with diverse agent algorithms to simulate rich multi-agent trajectories across 25+ economic tasks for AI-driven policy learning and analysis. Experiments show that EconGym supports diverse and cross-domain tasks-such as coordinating fiscal, pension, and monetary policies-and enables benchmarking across AI, economic methods, and hybrids. Results indicate that richer task composition and algorithm diversity expand the policy space, while AI agents guided by classical economic methods perform best in complex settings. EconGym also scales to 10k agents with high realism and efficiency.

econ.GN↗

arXiv2025

Learning Macroeconomic Policies through Dynamic Stackelberg Mean-Field Games

Macroeconomic outcomes emerge from individuals' decisions, making it essential to model how agents interact with macro policy via consumption, investment, and labor choices. We formulate this as a dynamic Stackelberg game: the government (leader) sets policies, and agents (followers) respond by optimizing their behavior over time. Unlike static models, this dynamic formulation captures temporal dependencies and strategic feedback critical to policy design. However, as the number of agents increases, explicitly simulating all agent-agent and agent-government interactions becomes computationally infeasible. To address this, we propose the Dynamic Stackelberg Mean Field Game (DSMFG) framework, which approximates these complex interactions via agent-population and government-population couplings. This approximation preserves individual-level feedback while ensuring scalability, enabling DSMFG to jointly model three core features of real-world policymaking: dynamic feedback, asymmetry, and large scale. We further introduce Stackelberg Mean Field Reinforcement Learning (SMFRL), a data-driven algorithm that learns the leader's optimal policies while maintaining personalized responses for individual agents. Empirically, we validate our approach in a large-scale simulated economy, where it scales to 1,000 agents (vs. 100 in prior work) and achieves a fourfold increase in GDP over classical economic methods and a nineteenfold improvement over the static 2022 U.S. federal income tax policy.

econ.TH↗

arXiv2025

UniMLVG: Unified Framework for Multi-view Long Video Generation with Comprehensive Control Capabilities for Autonomous Driving

The creation of diverse and realistic driving scenarios has become essential to enhance perception and planning capabilities of the autonomous driving system. However, generating long-duration, surround-view consistent driving videos remains a significant challenge. To address this, we present UniMLVG, a unified framework designed to generate extended street multi-perspective videos under precise control. By integrating single- and multi-view driving videos into the training data, our approach updates a DiT-based diffusion model equipped with cross-frame and cross-view modules across three stages with multi training objectives, substantially boosting the diversity and quality of generated visual content. Importantly, we propose an innovative explicit viewpoint modeling approach for multi-view video generation to effectively improve motion transition consistency. Capable of handling various input reference formats (e.g., text, images, or video), our UniMLVG generates high-quality multi-view videos according to the corresponding condition constraints such as 3D bounding boxes or frame-level text descriptions. Compared to the best models with similar capabilities, our framework achieves improvements of 48.2% in FID and 35.2% in FVD.

cs.CV↗

arXiv2024

EA-RAS: Towards Efficient and Accurate End-to-End Reconstruction of Anatomical Skeleton

Efficient, accurate and low-cost estimation of human skeletal information is crucial for a range of applications such as biology education and human-computer interaction. However, current simple skeleton models, which are typically based on 2D-3D joint points, fall short in terms of anatomical fidelity, restricting their utility in fields. On the other hand, more complex models while anatomically precise, are hindered by sophisticate multi-stage processing and the need for extra data like skin meshes, making them unsuitable for real-time applications. To this end, we propose the EA-RAS (Towards Efficient and Accurate End-to-End Reconstruction of Anatomical Skeleton), a single-stage, lightweight, and plug-and-play anatomical skeleton estimator that can provide real-time, accurate anatomically realistic skeletons with arbitrary pose using only a single RGB image input. Additionally, EA-RAS estimates the conventional human-mesh model explicitly, which not only enhances the functionality but also leverages the outside skin information by integrating features into the inside skeleton modeling process. In this work, we also develop a progressive training strategy and integrated it with an enhanced optimization process, enabling the network to obtain initial weights using only a small skin dataset and achieve self-supervision in skeleton reconstruction. Besides, we also provide an optional lightweight post-processing optimization strategy to further improve accuracy for scenarios that prioritize precision over real-time processing. The experiments demonstrated that our regression method is over 800 times faster than existing methods, meeting real-time requirements. Additionally, the post-processing optimization strategy provided can enhance reconstruction accuracy by over 50% and achieve a speed increase of more than 7 times.

cs.CV↗

arXiv2024

Cross-Block Fine-Grained Semantic Cascade for Skeleton-Based Sports Action Recognition

Human action video recognition has recently attracted more attention in applications such as video security and sports posture correction. Popular solutions, including graph convolutional networks (GCNs) that model the human skeleton as a spatiotemporal graph, have proven very effective. GCNs-based methods with stacked blocks usually utilize top-layer semantics for classification/annotation purposes. Although the global features learned through the procedure are suitable for the general classification, they have difficulty capturing fine-grained action change across adjacent frames -- decisive factors in sports actions. In this paper, we propose a novel ``Cross-block Fine-grained Semantic Cascade (CFSC)'' module to overcome this challenge. In summary, the proposed CFSC progressively integrates shallow visual knowledge into high-level blocks to allow networks to focus on action details. In particular, the CFSC module utilizes the GCN feature maps produced at different levels, as well as aggregated features from proceeding levels to consolidate fine-grained features. In addition, a dedicated temporal convolution is applied at each level to learn short-term temporal features, which will be carried over from shallow to deep layers to maximize the leverage of low-level details. This cross-block feature aggregation methodology, capable of mitigating the loss of fine-grained information, has resulted in improved performance. Last, FD-7, a new action recognition dataset for fencing sports, was collected and will be made publicly available. Experimental results and empirical analysis on public benchmarks (FSD-10) and self-collected (FD-7) demonstrate the advantage of our CFSC module on learning discriminative patterns for action classification over others.

cs.CV↗

arXiv2024

CSTalk: Correlation Supervised Speech-driven 3D Emotional Facial Animation Generation

Speech-driven 3D facial animation technology has been developed for years, but its practical application still lacks expectations. The main challenges lie in data limitations, lip alignment, and the naturalness of facial expressions. Although lip alignment has seen many related studies, existing methods struggle to synthesize natural and realistic expressions, resulting in a mechanical and stiff appearance of facial animations. Even with some research extracting emotional features from speech, the randomness of facial movements limits the effective expression of emotions. To address this issue, this paper proposes a method called CSTalk (Correlation Supervised) that models the correlations among different regions of facial movements and supervises the training of the generative model to generate realistic expressions that conform to human facial motion patterns. To generate more intricate animations, we employ a rich set of control parameters based on the metahuman character model and capture a dataset for five different emotions. We train a generative network using an autoencoder structure and input an emotion embedding vector to achieve the generation of user-control expressions. Experimental results demonstrate that our method outperforms existing state-of-the-art methods.

cs.CV↗

arXiv2024

Embedded Representation Learning Network for Animating Styled Video Portrait

The talking head generation recently attracted considerable attention due to its widespread application prospects, especially for digital avatars and 3D animation design. Inspired by this practical demand, several works explored Neural Radiance Fields (NeRF) to synthesize the talking heads. However, these methods based on NeRF face two challenges: (1) Difficulty in generating style-controllable talking heads. (2) Displacement artifacts around the neck in rendered images. To overcome these two challenges, we propose a novel generative paradigm \textit{Embedded Representation Learning Network} (ERLNet) with two learning stages. First, the \textit{ audio-driven FLAME} (ADF) module is constructed to produce facial expression and head pose sequences synchronized with content audio and style video. Second, given the sequence deduced by the ADF, one novel \textit{dual-branch fusion NeRF} (DBF-NeRF) explores these contents to render the final images. Extensive empirical studies demonstrate that the collaboration of these two stages effectively facilitates our method to render a more realistic talking head than the existing algorithms.

cs.CV↗

arXiv2024

Sketch3D: Style-Consistent Guidance for Sketch-to-3D Generation

Recently, image-to-3D approaches have achieved significant results with a natural image as input. However, it is not always possible to access these enriched color input samples in practical applications, where only sketches are available. Existing sketch-to-3D researches suffer from limitations in broad applications due to the challenges of lacking color information and multi-view content. To overcome them, this paper proposes a novel generation paradigm Sketch3D to generate realistic 3D assets with shape aligned with the input sketch and color matching the textual description. Concretely, Sketch3D first instantiates the given sketch in the reference image through the shape-preserving generation process. Second, the reference image is leveraged to deduce a coarse 3D Gaussian prior, and multi-view style-consistent guidance images are generated based on the renderings of the 3D Gaussians. Finally, three strategies are designed to optimize 3D Gaussians, i.e., structural optimization via a distribution transfer mechanism, color optimization with a straightforward MSE loss and sketch similarity optimization with a CLIP-based geometric similarity loss. Extensive visual comparisons and quantitative analysis illustrate the advantage of our Sketch3D in generating realistic 3D assets while preserving consistency with the input.

cs.CV↗

arXiv2024

TaxAI: A Dynamic Economic Simulator and Benchmark for Multi-Agent Reinforcement Learning

Taxation and government spending are crucial tools for governments to promote economic growth and maintain social equity. However, the difficulty in accurately predicting the dynamic strategies of diverse self-interested households presents a challenge for governments to implement effective tax policies. Given its proficiency in modeling other agents in partially observable environments and adaptively learning to find optimal policies, Multi-Agent Reinforcement Learning (MARL) is highly suitable for solving dynamic games between the government and numerous households. Although MARL shows more potential than traditional methods such as the genetic algorithm and dynamic programming, there is a lack of large-scale multi-agent reinforcement learning economic simulators. Therefore, we propose a MARL environment, named \textbf{TaxAI}, for dynamic games involving $N$ households, government, firms, and financial intermediaries based on the Bewley-Aiyagari economic model. Our study benchmarks 2 traditional economic methods with 7 MARL methods on TaxAI, demonstrating the effectiveness and superiority of MARL algorithms. Moreover, TaxAI's scalability in simulating dynamic interactions between the government and 10,000 households, coupled with real-data calibration, grants it a substantial improvement in scale and reality over existing simulators. Therefore, TaxAI is the most realistic economic simulator for optimal tax policy, which aims to generate feasible recommendations for governments and individuals.

cs.CE↗

arXiv2023

Few-shot Shape Recognition by Learning Deep Shape-aware Features

Traditional shape descriptors have been gradually replaced by convolutional neural networks due to their superior performance in feature extraction and classification. The state-of-the-art methods recognize object shapes via image reconstruction or pixel classification. However , these methods are biased toward texture information and overlook the essential shape descriptions, thus, they fail to generalize to unseen shapes. We are the first to propose a fewshot shape descriptor (FSSD) to recognize object shapes given only one or a few samples. We employ an embedding module for FSSD to extract transformation-invariant shape features. Secondly, we develop a dual attention mechanism to decompose and reconstruct the shape features via learnable shape primitives. In this way, any shape can be formed through a finite set basis, and the learned representation model is highly interpretable and extendable to unseen shapes. Thirdly, we propose a decoding module to include the supervision of shape masks and edges and align the original and reconstructed shape features, enforcing the learned features to be more shape-aware. Lastly, all the proposed modules are assembled into a few-shot shape recognition scheme. Experiments on five datasets show that our FSSD significantly improves the shape classification compared to the state-of-the-art under the few-shot setting.

cs.CV↗

arXiv2022

Unsupervised Visual Defect Detection with Score-Based Generative Model

Anomaly Detection (AD), as a critical problem, has been widely discussed. In this paper, we specialize in one specific problem, Visual Defect Detection (VDD), in many industrial applications. And in practice, defect image samples are very rare and difficult to collect. Thus, we focus on the unsupervised visual defect detection and localization tasks and propose a novel framework based on the recent score-based generative models, which synthesize the real image by iterative denoising through stochastic differential equations (SDEs). Our work is inspired by the fact that with noise injected into the original image, the defects may be changed into normal cases in the denoising process (i.e., reconstruction). First, based on the assumption that the anomalous data lie in the low probability density region of the normal data distribution, we explain a common phenomenon that occurs when reconstruction-based approaches are applied to VDD: normal pixels also change during the reconstruction process. Second, due to the differences in normal pixels between the reconstructed and original images, a time-dependent gradient value (i.e., score) of normal data distribution is utilized as a metric, rather than reconstruction loss, to gauge the defects. Third, a novel $T$ scales approach is developed to dramatically reduce the required number of iterations, accelerating the inference process. These practices allow our model to generalize VDD in an unsupervised manner while maintaining reasonably good performance. We evaluate our method on several datasets to demonstrate its effectiveness.

cs.CV↗

arXiv2020

Towards 3D Dance Motion Synthesis and Control

3D human dance motion is a cooperative and elegant social movement. Unlike regular simple locomotion, it is challenging to synthesize artistic dance motions due to the irregularity, kinematic complexity and diversity. It requires the synthesized dance is realistic, diverse and controllable. In this paper, we propose a novel generative motion model based on temporal convolution and LSTM,TC-LSTM, to synthesize realistic and diverse dance motion. We introduce a unique control signal, dance melody line, to heighten controllability. Hence, our model, and its switch for control signals, promote a variety of applications: random dance synthesis, music-to-dance, user control, and more. Our experiments demonstrate that our model can synthesize artistic dance motion in various dance types. Compared with existing methods, our method achieved start-of-the-art results.

cs.GR↗