Bin Lei

Le Chen, Nuo Xu, Winson Chen et al.

Large language models (LLMs) have shown remarkable capabilities in code translation, yet their performance deteriorates in low-resource programming domains such as Fortran and emerging frameworks like CUDA, where high-quality parallel data are scarce. We present an automated dataset generation pipeline featuring a dual-LLM Questioner-Solver design that incorporates external knowledge from compilers and runtime feedback. Beyond traditional source-target code pair datasets, our approach additionally generates (1) verified translations with unit tests for assessing functional consistency and (2) multi-turn dialogues that capture the reasoning process behind translation refinement. Applied to Fortran-to-C++ and C++-to-CUDA, the pipeline yields 3.64k and 3.93k dialogues, respectively. Fine-tuning on this data yields dramatic improvements in functional correctness, boosting unit test success rates by over 56% on the challenging C++-to-CUDA task. We show that the generated data enables a 7B open-weight model to significantly outperform larger proprietary systems on key metrics like compilation success.

Bin Lei, Caiwen Ding, Le Chen et al.

In this study, we present a novel dataset for training machine learning models translating between OpenMP Fortran and C++ code. To ensure reliability and applicability, the dataset is created from a range of representative open-source OpenMP benchmarks. It is also refined using a meticulous code similarity test. The effectiveness of our dataset is assessed using both quantitative (CodeBLEU) and qualitative (human evaluation) methods. We showcase how this dataset significantly elevates the translation competencies of large language models (LLMs). Specifically, models without prior coding knowledge experienced a boost of $\mathbf{\times~5.1}$ in their CodeBLEU scores, while models with some coding familiarity saw an impressive $\mathbf{\times~9.9}$-fold increase. The best fine-tuned model using our dataset outperforms GPT-4. It is also reaching human-level accuracy. This work underscores the immense potential of our dataset in propelling advancements in the domain of code translation for high-performance computing. The dataset is accessible at \href{https://github.com/bin123apple/Fortran-CPP-HPC-code-translation-dataset}{OpenMP-Fortran-CPP-Translation}.

Bin Lei, Weitai Kang, Zijian Zhang et al.

This paper introduces \textsc{InfantAgent-Next}, a generalist agent capable of interacting with computers in a multimodal manner, encompassing text, images, audio, and video. Unlike existing approaches that either build intricate workflows around a single large model or only provide workflow modularity, our agent integrates tool-based and pure vision agents within a highly modular architecture, enabling different models to collaboratively solve decoupled tasks in a step-by-step manner. Our generality is demonstrated by our ability to evaluate not only pure vision-based real-world benchmarks (i.e., OSWorld), but also more general or tool-intensive benchmarks (e.g., GAIA and SWE-Bench). Specifically, we achieve $\mathbf{7.27\%}$ accuracy on OSWorld, higher than Claude-Computer-Use. Codes and evaluation scripts are open-sourced at https://github.com/bin123apple/InfantAgent.

Le Chen, Arijit Bhattacharjee, Nesreen K. Ahmed et al.

Large language models (LLMs) have become increasingly prominent in academia and industry due to their remarkable performance in diverse applications. As these models evolve with increasing parameters, they excel in tasks like sentiment analysis and machine translation. However, even models with billions of parameters face challenges in tasks demanding multi-step reasoning. Code generation and comprehension, especially in C and C++, emerge as significant challenges. While LLMs trained on code datasets demonstrate competence in many tasks, they struggle with rectifying non-compilable C and C++ code. Our investigation attributes this subpar performance to two primary factors: the quality of the training dataset and the inherent complexity of the problem which demands intricate reasoning. Existing "Chain of Thought" (CoT) prompting techniques aim to enhance multi-step reasoning. This approach, however, retains the limitations associated with the latent drawbacks of LLMs. In this work, we propose CompCodeVet, a compiler-guided CoT approach to produce compilable code from non-compilable ones. Diverging from the conventional approach of utilizing larger LLMs, we employ compilers as a teacher to establish a more robust zero-shot thought process. The evaluation of CompCodeVet on two open-source code datasets shows that CompCodeVet has the ability to improve the training dataset quality for LLMs.

Bin Lei, pei-Hung Lin, Chunhua Liao et al.

Recent advancements in large-scale models, such as GPT-4, have showcased remarkable capabilities in addressing standard queries. However, when facing complex problems that require multi-step logical reasoning, their accuracy dramatically decreases. Current research has explored the realm of \textit{prompting engineering} to bolster the inferential capacities of these models. Our paper unveils a pioneering prompting technique, dubbed \textit{Graph of Thoughts (GoT)}. Through testing on a trio of escalating challenges: the 24-point game, resolution of high-degree polynomial equations, and derivation of formulas for recursive sequences, our method outperformed GPT-4, achieving accuracy improvements of $89.7\%$, $86\%$, and $56\%$ for each respective task. Moreover, when juxtaposed with the state-of-the-art (SOTA) prompting method, \textit{Tree of Thought (ToT)}, our approach registered an average accuracy boost of $23\%$, $24\%$, and $15\%$.

Bin Lei, Shaoyi Huang, Caiwen Ding et al.

Traffic state prediction in a transportation network is paramount for effective traffic operations and management, as well as informed user and system-level decision-making. However, long-term traffic prediction (beyond 30 minutes into the future) remains challenging in current research. In this work, we integrate the spatio-temporal dependencies in the transportation network from network modeling, together with the graph convolutional network (GCN) and graph attention network (GAT). To further tackle the dramatic computation and memory cost caused by the giant model size (i.e., number of weights) caused by multiple cascaded layers, we propose sparse training to mitigate the training cost, while preserving the prediction accuracy. It is a process of training using a fixed number of nonzero weights in each layer in each iteration. We consider the problem of long-term traffic speed forecasting for a real large-scale transportation network data from the California Department of Transportation (Caltrans) Performance Measurement System (PeMS). Experimental results show that the proposed GCN-STGT and GAT-STGT models achieve low prediction errors on short-, mid- and long-term prediction horizons, of 15, 30 and 45 minutes in duration, respectively. Using our sparse training, we could train from scratch with high sparsity (e.g., up to 90%), equivalent to 10 times floating point operations per second (FLOPs) reduction on computational cost using the same epochs as dense training, and arrive at a model with very small accuracy loss compared with the original dense training

Bin Lei, Sheng Lin, Pei-Hung Lin et al.

Biologically-inspired Spiking Neural Networks (SNNs), processing information using discrete-time events known as spikes rather than continuous values, have garnered significant attention due to their hardware-friendly and energy-efficient characteristics. However, the training of SNNs necessitates a considerably large memory footprint, given the additional storage requirements for spikes or events, leading to a complex structure and dynamic setup. In this paper, to address memory constraint in SNN training, we introduce an innovative framework, characterized by a remarkably low memory footprint. We \textbf{(i)} design a reversible SNN node that retains a high level of accuracy. Our design is able to achieve a $\mathbf{58.65\times}$ reduction in memory usage compared to the current SNN node. We \textbf{(ii)} propose a unique algorithm to streamline the backpropagation process of our reversible SNN node. This significantly trims the backward Floating Point Operations Per Second (FLOPs), thereby accelerating the training process in comparison to current reversible layer backpropagation method. By using our algorithm, the training time is able to be curtailed by $\mathbf{23.8\%}$ relative to existing reversible layer architectures.

Bin Lei, Yi Zhang, Shan Zuo et al.

Recent advancements in large language models, such as GPT-4, have demonstrated remarkable capabilities in processing standard queries. Despite these advancements, their performance substantially declines in \textbf{advanced mathematical problems requiring complex, multi-step logical reasoning}. To enhance their inferential capabilities, current research has delved into \textit{prompting engineering}, exemplified by methodologies such as the Tree of Thought and Graph of Thought. Nonetheless, these existing approaches encounter two significant limitations. Firstly, their effectiveness in tackling complex mathematical problems is somewhat constrained. Secondly, the necessity to design distinct prompts for individual problems hampers their generalizability. In response to these limitations, this paper introduces the \textit{Multi-Agent System for conditional Mining} (\textbf{MACM}) prompting method. It not only resolves intricate mathematical problems but also demonstrates strong generalization capabilities across various mathematical contexts. With the assistance of MACM, the accuracy of GPT-4 Turbo on the most challenging level five mathematical problems in the MATH dataset increase from $\mathbf{54.68\%} \text{ to } \mathbf{76.73\%}$. The code is available in \url{https://github.com/bin123apple/MACM}.

Bin Lei, Yuchen Li, Qiuwu Chen

We introduce AutoCoder, the first Large Language Model to surpass GPT-4 Turbo (April 2024) and GPT-4o in pass@1 on the Human Eval benchmark test ($\mathbf{90.9\%}$ vs. $\mathbf{90.2\%}$). In addition, AutoCoder offers a more versatile code interpreter compared to GPT-4 Turbo and GPT-4o. It's code interpreter can install external packages instead of limiting to built-in packages. AutoCoder's training data is a multi-turn dialogue dataset created by a system combining agent interaction and external code execution verification, a method we term \textbf{\textsc{AIEV-Instruct}} (Instruction Tuning with Agent-Interaction and Execution-Verified). Compared to previous large-scale code dataset generation methods, \textsc{AIEV-Instruct} reduces dependence on proprietary large models and provides execution-validated code dataset. The code and the demo video is available in \url{https://github.com/bin123apple/AutoCoder}.

Zeyuan Li, Yangfan He, Lewei He et al.

Recently, large language models (LLMs) have achieved significant progress in automated code generation. Despite their strong instruction-following capabilities, these models frequently struggled to align with user intent in coding scenarios. In particular, they were hampered by datasets that lacked diversity and failed to address specialized tasks or edge cases. Furthermore, challenges in supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF) led to failures in generating precise, human-intent-aligned code. To tackle these challenges and improve the code generation performance for automated programming systems, we propose Feedback-driven Adaptive Long/short-term memory reinforced Coding Optimization (i.e., FALCON). FALCON is structured into two hierarchical levels. From the global level, long-term memory improves code quality by retaining and applying learned knowledge. At the local level, short-term memory allows for the incorporation of immediate feedback from compilers and AI systems. Additionally, we introduce meta-reinforcement learning with feedback rewards to solve the global-local bi-level optimization problem and enhance the model's adaptability across diverse code generation tasks. Extensive experiments demonstrate that our technique achieves state-of-the-art performance, leading other reinforcement learning methods by more than 4.5 percentage points on the MBPP benchmark and 6.1 percentage points on the Humaneval benchmark. The open-sourced code is publicly available at https://github.com/titurte/FALCON.

Le Chen, Bin Lei, Dunzhi Zhou et al.

Translating legacy Fortran code into C++ is a crucial step in modernizing high-performance computing (HPC) applications. However, the scarcity of high-quality, parallel Fortran-to-C++ datasets and the limited domain-specific expertise in large language models (LLMs) present significant challenges for automated translation. In this paper, we introduce Fortran2CPP, a multi-turn dialogue dataset generated by a novel LLM agent-based approach that integrates a dual-LLM Questioner-Solver module to enhance translation accuracy. Our dataset comprises 11.7k dialogues capturing iterative feedback-decision workflows including code translation, compilation, execution, unit testing, and error-fixing. Using this dataset, we fine-tune several open-weight LLMs and achieve up to a 3.31x improvement in CodeBLEU scores and a 92\% increase in compilation success rate, demonstrating enhanced syntactic accuracy and functional reliability. Our findings highlight the value of dialogue-based LLM training for complex code translation tasks. The dataset and model have been open-sourced and are available on our public GitHub repository\footnote{\url{https://github.com/HPC-Fortran2CPP/Fortran2Cpp}}.

Bin Lei, Caiwen Ding, Jiachen Yang et al.

Chain-of-Thought (CoT) reasoning has become a foundation for eliciting multi-step reasoning in large language models, but recent studies show that its benefits do not scale monotonically with chain length: while longer CoT generally enables a model to tackle harder problems, on a given problem, accuracy typically increases with CoT length up to a point, after which it declines. We identify a major cause of this phenomenon: as the CoT grows, the model's attention to critical insights produced earlier in the trace gradually weakens, making those insights progressively less accessible when they are most needed. Therefore, we propose \textbf{InsightReplay}, a stateful reasoning approach in which the model periodically extracts critical insights from its reasoning trace and replays them near the active generation frontier, keeping them accessible as the reasoning scales. Extensive experiments on a $\mathbf{2}\!\times\!\mathbf{3}\!\times\!\mathbf{4}$ benchmark grid, covering model scales $\{\text{8B}, \text{30B}\}$, model families $\{\text{Qwen3.5}, \text{DeepSeek-R1-Distill-Qwen}, \text{Gemma-4}\}$, and reasoning benchmarks $\{\text{AIME}, \text{HMMT}, \text{GPQA Diamond}, \text{LiveCodeBench v5}\}$, show that 3-round InsightReplay yields accuracy gains across \textbf{all 24 settings}, with an averaged improvement of $\mathbf{+1.65}$ points over standard CoT, and a largest single-setting gain of $\mathbf{+9.2}$ points on R1-Distill-32B's LiveCodeBench v5 subset. Our results suggest that the effectiveness of test-time scaling depends not only on how much a model reasons, but also on whether critical intermediate insights remain accessible throughout long reasoning trajectories.

Lei Liu, Zongxu Pan, Bin Lei

Detection of arbitrarily rotated objects is a challenging task due to the difficulties of locating the multi-angle objects and separating them effectively from the background. The existing methods are not robust to angle varies of the objects because of the use of traditional bounding box, which is a rotation variant structure for locating rotated objects. In this article, a new detection method is proposed which applies the newly defined rotatable bounding box (RBox). The proposed detector (DRBox) can effectively handle the situation where the orientation angles of the objects are arbitrary. The training of DRBox forces the detection networks to learn the correct orientation angle of the objects, so that the rotation invariant property can be achieved. DRBox is tested to detect vehicles, ships and airplanes on satellite images, compared with Faster R-CNN and SSD, which are chosen as the benchmark of the traditional bounding box based methods. The results shows that DRBox performs much better than traditional bounding box based methods do on the given tasks, and is more robust against rotation of input image and target objects. Besides, results show that DRBox correctly outputs the orientation angles of the objects, which is very useful for locating multi-angle objects efficiently. The code and models are available at https://github.com/liulei01/DRBox.

Bin Lei, Yuchen Li, Yiming Zeng et al.

Despite the impressive capabilities of large language models (LLMs), they currently exhibit two primary limitations, \textbf{\uppercase\expandafter{\romannumeral 1}}: They struggle to \textbf{autonomously solve the real world engineering problem}. \textbf{\uppercase\expandafter{\romannumeral 2}}: They remain \textbf{challenged in reasoning through complex logic problems}. To address these challenges, we developed the \textsc{Infant Agent}, integrating task-aware functions, operators, a hierarchical management system, and a memory retrieval mechanism. Together, these components enable large language models to sustain extended reasoning processes and handle complex, multi-step tasks efficiently, all while significantly reducing API costs. Using the \textsc{Infant Agent}, GPT-4o's accuracy on the SWE-bench-lite dataset rises from $\mathbf{0.33\%}$ to $\mathbf{30\%}$, and in the AIME-2024 mathematics competition, it increases GPT-4o's accuracy from $\mathbf{13.3\%}$ to $\mathbf{37\%}$.

Bin Lei, le Chen, Caiwen Ding

In the evolving field of machine learning, video generation has witnessed significant advancements with autoregressive-based transformer models and diffusion models, known for synthesizing dynamic and realistic scenes. However, these models often face challenges with prolonged inference times, even for generating short video clips such as GIFs. This paper introduces FlashVideo, a novel framework tailored for swift Text-to-Video generation. FlashVideo represents the first successful adaptation of the RetNet architecture for video generation, bringing a unique approach to the field. Leveraging the RetNet-based architecture, FlashVideo reduces the time complexity of inference from $\mathcal{O}(L^2)$ to $\mathcal{O}(L)$ for a sequence of length $L$, significantly accelerating inference speed. Additionally, we adopt a redundant-free frame interpolation method, enhancing the efficiency of frame interpolation. Our comprehensive experiments demonstrate that FlashVideo achieves a $\times9.17$ efficiency improvement over a traditional autoregressive-based transformer model, and its inference speed is of the same order of magnitude as that of BERT-based transformer models.

Weitai Kang, Bin Lei, Gaowen Liu et al.

Graphical user interface visual grounding (GUI-VG), a core capability for GUI agents, has primarily relied on supervised fine-tuning (SFT) of multimodal large language models (MLLMs), which demands extensive data curation and significant training costs. However, as MLLMs continue to advance and even cover GUI domains during pretraining, the necessity of exhaustive SFT post-training becomes increasingly questionable. Meanwhile, recent successes of rule-based reinforcement fine-tuning (RFT) suggest a more efficient alternative. Despite this promise, the optimal manner of applying RFT for GUI-VG remains unexplored. To bridge this gap, we introduce GuirlVG, a reinforcement learning-based GUI-VG method built on a systematic empirical study and a novel stabilization technique. We find that naive application of RFT underperforms the SFT baseline, motivating a deeper exploration. First, we decompose RFT into its core components and analyze the optimal formulation of each. Second, we propose a novel Adversarial KL Factor that dynamically stabilizes training to mitigate reward over-optimization. Third, we further explore the training configurations of RFT to enhance effectiveness. Extensive experiments show that GuirlVG, with only 5.2K training samples, outperforms SFT methods trained on over 10M samples, achieving a 7.7% improvement on ScreenSpot, a 17.2% improvement on ScreenSpotPro, and 91.9% accuracy on ScreenSpotV2.

Bin Lei, Nuo Xu, Ali Payani et al.

Multimodal large language models (MLLMs) have markedly expanded the competence of graphical user-interface (GUI) systems, propelling them beyond controlled simulations into complex, real-world environments across diverse platforms. However, practical usefulness is still bounded by the reliability of visual grounding, i.e., mapping textual references to exact on-screen elements. This limitation prevents the system from accurately performing pointer-level actions such as clicking or dragging. To address it, we introduce GUI-Spotlight -- a model trained for image-grounded reasoning that dynamically invokes multiple specialized tools to iteratively narrow its focus to the relevant region of the screen, thereby substantially improving visual grounding accuracy. On the ScreenSpot-Pro benchmark, GUI-Spotlight trained with only 18.5K training samples achieves 52.8\% accuracy, surpassing V2P-7B (50.6\% with 9.6M training samples) and GTA-1-7B (50.1\% with 1.56M training samples).

Kiran Thorat, Jiahui Zhao, Yaotian Liu et al.

Large Language Models (LLMs) are gaining prominence in various fields, thanks to their ability to generate high- quality content from human instructions. This paper delves into the field of chip design using LLMs, specifically in Power- Performance-Area (PPA) optimization and the generation of accurate Verilog codes for circuit designs. We introduce a novel framework VeriPPA designed to optimize PPA and generate Verilog code using LLMs. Our method includes a two-stage process where the first stage focuses on improving the functional and syntactic correctness of the generated Verilog codes, while the second stage focuses on optimizing the Verilog codes to meet PPA constraints of circuit designs, a crucial element of chip design. Our framework achieves an 81.37% success rate in syntactic correctness and 62.06% in functional correctness for code genera- tion, outperforming current state-of-the-art (SOTA) methods. On the RTLLM dataset. On the VerilogEval dataset, our framework achieves 99.56% syntactic correctness and 43.79% functional correctness, also surpassing SOTA, which stands at 92.11% for syntactic correctness and 33.57% for functional correctness. Furthermore, Our framework able to optimize the PPA of the designs. These results highlight the potential of LLMs in handling complex technical areas and indicate an encouraging development in the automation of chip design processes.

Zhongling Huang, Corneliu Octavian Dumitru, Zongxu Pan et al.

The classification of large-scale high-resolution SAR land cover images acquired by satellites is a challenging task, facing several difficulties such as semantic annotation with expertise, changing data characteristics due to varying imaging parameters or regional target area differences, and complex scattering mechanisms being different from optical imaging. Given a large-scale SAR land cover dataset collected from TerraSAR-X images with a hierarchical three-level annotation of 150 categories and comprising more than 100,000 patches, three main challenges in automatically interpreting SAR images of highly imbalanced classes, geographic diversity, and label noise are addressed. In this letter, a deep transfer learning method is proposed based on a similarly annotated optical land cover dataset (NWPU-RESISC45). Besides, a top-2 smooth loss function with cost-sensitive parameters was introduced to tackle the label noise and imbalanced classes' problems. The proposed method shows high efficiency in transferring information from a similarly annotated remote sensing dataset, a robust performance on highly imbalanced classes, and is alleviating the over-fitting problem caused by label noise. What's more, the learned deep model has a good generalization for other SAR-specific tasks, such as MSTAR target recognition with a state-of-the-art classification accuracy of 99.46%.

Zhongling Huang, Zongxu Pan, Bin Lei

Deep convolutional neural networks (DCNNs) have attracted much attention in remote sensing recently. Compared with the large-scale annotated dataset in natural images, the lack of labeled data in remote sensing becomes an obstacle to train a deep network very well, especially in SAR image interpretation. Transfer learning provides an effective way to solve this problem by borrowing the knowledge from the source task to the target task. In optical remote sensing application, a prevalent mechanism is to fine-tune on an existing model pre-trained with a large-scale natural image dataset, such as ImageNet. However, this scheme does not achieve satisfactory performance for SAR application because of the prominent discrepancy between SAR and optical images. In this paper, we attempt to discuss three issues that are seldom studied before in detail: (1) what network and source tasks are better to transfer to SAR targets, (2) in which layer are transferred features more generic to SAR targets and (3) how to transfer effectively to SAR targets recognition. Based on the analysis, a transitive transfer method via multi-source data with domain adaptation is proposed in this paper to decrease the discrepancy between the source data and SAR targets. Several experiments are conducted on OpenSARShip. The results indicate that the universal conclusions about transfer learning in natural images cannot be completely applied to SAR targets, and the analysis of what and where to transfer in SAR target recognition is helpful to decide how to transfer more effectively.