Yinheng Li

Rogerio Bonatti, Dan Zhao, Francesco Bonacci et al.

Large language models (LLMs) show remarkable potential to act as computer agents, enhancing human productivity and software accessibility in multi-modal tasks that require planning and reasoning. However, measuring agent performance in realistic environments remains a challenge since: (i) most benchmarks are limited to specific modalities or domains (e.g. text-only, web navigation, Q&A, coding) and (ii) full benchmark evaluations are slow (on order of magnitude of days) given the multi-step sequential nature of tasks. To address these challenges, we introduce the Windows Agent Arena: a reproducible, general environment focusing exclusively on the Windows operating system (OS) where agents can operate freely within a real Windows OS and use the same wide range of applications, tools, and web browsers available to human users when solving tasks. We adapt the OSWorld framework (Xie et al., 2024) to create 150+ diverse Windows tasks across representative domains that require agent abilities in planning, screen understanding, and tool usage. Our benchmark is scalable and can be seamlessly parallelized in Azure for a full benchmark evaluation in as little as 20 minutes. To demonstrate Windows Agent Arena's capabilities, we also introduce a new multi-modal agent, Navi. Our agent achieves a success rate of 19.5% in the Windows domain, compared to 74.5% performance of an unassisted human. Navi also demonstrates strong performance on another popular web-based benchmark, Mind2Web. We offer extensive quantitative and qualitative analysis of Navi's performance, and provide insights into the opportunities for future research in agent development and data generation using Windows Agent Arena. Webpage: https://microsoft.github.io/WindowsAgentArena Code: https://github.com/microsoft/WindowsAgentArena

Yinheng Li, Shaofei Wang, Han Ding et al.

Recent advances in large language models (LLMs) have opened new possibilities for artificial intelligence applications in finance. In this paper, we provide a practical survey focused on two key aspects of utilizing LLMs for financial tasks: existing solutions and guidance for adoption. First, we review current approaches employing LLMs in finance, including leveraging pretrained models via zero-shot or few-shot learning, fine-tuning on domain-specific data, and training custom LLMs from scratch. We summarize key models and evaluate their performance improvements on financial natural language processing tasks. Second, we propose a decision framework to guide financial professionals in selecting the appropriate LLM solution based on their use case constraints around data, compute, and performance needs. The framework provides a pathway from lightweight experimentation to heavy investment in customized LLMs. Lastly, we discuss limitations and challenges around leveraging LLMs in financial applications. Overall, this survey aims to synthesize the state-of-the-art and provide a roadmap for responsibly applying LLMs to advance financial AI.

Yinheng Li

The remarkable advancements in large language models (LLMs) have brought about significant improvements in Natural Language Processing(NLP) tasks. This paper presents a comprehensive review of in-context learning techniques, focusing on different types of prompts, including discrete, continuous, few-shot, and zero-shot, and their impact on LLM performance. We explore various approaches to prompt design, such as manual design, optimization algorithms, and evaluation methods, to optimize LLM performance across diverse tasks. Our review covers key research studies in prompt engineering, discussing their methodologies and contributions to the field. We also delve into the challenges faced in evaluating prompt performance, given the absence of a single "best" prompt and the importance of considering multiple metrics. In conclusion, the paper highlights the critical role of prompt design in harnessing the full potential of LLMs and provides insights into the combination of manual design, optimization techniques, and rigorous evaluation for more effective and efficient use of LLMs in various NLP tasks.

Han Ding, Yinheng Li, Junhao Wang et al.

Trading is a highly competitive task that requires a combination of strategy, knowledge, and psychological fortitude. With the recent success of large language models(LLMs), it is appealing to apply the emerging intelligence of LLM agents in this competitive arena and understanding if they can outperform professional traders. In this survey, we provide a comprehensive review of the current research on using LLMs as agents in financial trading. We summarize the common architecture used in the agent, the data inputs, and the performance of LLM trading agents in backtesting as well as the challenges presented in these research. This survey aims to provide insights into the current state of LLM-based financial trading agents and outline future research directions in this field.

Zheng Hui, Yinheng Li, Dan zhao et al.

Graphical User Interface (GUI) tasks are vital for automating workflows such as software testing, user interface navigation. For users, the GUI is the most intuitive platform for interacting with a computer. Previous work identified a key challenge in developing visual GUI agents: GUI grounding - the ability to accurately locate screen elements based on instructions. However, most existing GUI agents rely on structured data formats like DOM or HTML files in training or inferencing, which are inaccessible across all applications, particular in a general desktop environments such as Windows OS. To address this, we introduce WinClick, a novel visual GUI agent developed in Windows platform. WinClick leverages screenshots to detect actionable regions. To overcome the challenge of GUI grounding, we enhance WinClick with GUI grounding pre-training and propose an LLM-based method for aligning GUI grounding data. Additionally, we introduce WinSpot, the first comprehensive benchmark for GUI grounding on Windows. Our experiments demonstrate that WinClick, combined with GUI grounding pre-training, significantly outperforms existing baselines, offering a scalable solution for GUI automation in desktop environments. WinSpot is publicly available at https://github.com/zackhuiiiii/WinSpot.

Yinheng Li, Han Ding, Hang Chen

Data processing plays an significant role in current multimodal model training. In this paper. we provide an comprehensive review of common data processing techniques used in modern multimodal model training with a focus on diffusion models and multimodal large language models (MLLMs). We summarized all techniques into four categories: data quality, data quantity, data distribution and data safety. We further present our findings in the choice of data process methods in different type of models. This study aims to provide guidance to multimodal models developers with effective data processing techniques.

Zheng Hui, Zhaoxiao Guo, Hang Zhao et al.

Effective toxic content detection relies heavily on high-quality and diverse data, which serve as the foundation for robust content moderation models. Synthetic data has become a common approach for training models across various NLP tasks. However, its effectiveness remains uncertain for highly subjective tasks like hate speech detection, with previous research yielding mixed results. This study explores the potential of open-source LLMs for harmful data synthesis, utilizing controlled prompting and supervised fine-tuning techniques to enhance data quality and diversity. We systematically evaluated 6 open source LLMs on 5 datasets, assessing their ability to generate diverse, high-quality harmful data while minimizing hallucination and duplication. Our results show that Mistral consistently outperforms other open models, and supervised fine-tuning significantly enhances data reliability and diversity. We further analyze the trade-offs between prompt-based vs. fine-tuned toxic data synthesis, discuss real-world deployment challenges, and highlight ethical considerations. Our findings demonstrate that fine-tuned open source LLMs provide scalable and cost-effective solutions to augment toxic content detection datasets, paving the way for more accessible and transparent content moderation tools.

Tianyi Chen, Michael Solodko, Sen Wang et al.

Computer Using Agents (CUAs) are increasingly equipped with external tools, enabling them to perform complex and realistic tasks. For CUAs to operate effectively, application selection, which refers to deciding which application to use before invoking fine-grained tools such as APIs, is a fundamental capability. It determines whether the agent initializes the correct environment, avoids orchestration confusion, and efficiently focuses on relevant context. However, existing benchmarks primarily assess fine-grained API selection, offering limited insight into whether models can reason across and choose between different applications. To fill this gap, we introduce AppSelectBench, a comprehensive benchmark for evaluating application selection in CUAs. AppSelectBench contains a novel user task generation pipeline that produces realistic, diverse, and semantically grounded user intents at scale, together with unified evaluation protocols covering random, heuristic, zero-shot, few-shot, and retrieval-augmented-settings. AppSelectBench covers one hundred widely used desktop applications and includes more than one hundred thousand realistic, diverse, and semantically grounded user tasks. Extensive experiments across both closed-source and open-source large language models reveal systematic strengths and weaknesses in inter-application reasoning, showing that even the most capable models still struggle to make consistent application choices. Together, these results establish AppSelectBench as a foundation for studying and advancing application level reasoning, an essential yet underexplored capability of intelligent CUAs. The source is available at https://microsoft.github.io/appselectbench/.

Kewei Li, Yuqian Wu, Yinheng Li et al.

Since the mechanism of action of drug molecules in the human body is difficult to reproduce in the in vitro environment, it becomes difficult to reveal the causes of the activity cliff phenomenon of drug molecules. We found out the AC of small molecules has been extensively investigated but limited knowledge is accumulated about the AC phenomenon in peptides with canonical amino acids. Understanding the mechanism of AC in canonical amino acids might help understand the one in drug molecules. This study introduces a quantitative definition and benchmarking framework AMPCliff for the AC phenomenon in antimicrobial peptides (AMPs) composed by canonical amino acids. A comprehensive analysis of the existing AMP dataset reveals a significant prevalence of AC within AMPs. AMPCliff quantifies the activities of AMPs by the MIC, and defines 0.9 as the minimum threshold for the normalized BLOSUM62 similarity score between a pair of aligned peptides with at least two-fold MIC changes. This study establishes a benchmark dataset of paired AMPs in Staphylococcus aureus from the publicly available AMP dataset GRAMPA, and conducts a rigorous procedure to evaluate various AMP AC prediction models, including nine machine learning, four deep learning algorithms, four masked language models, and four generative language models. Our analysis reveals that these models are capable of detecting AMP AC events and the pre-trained protein language model ESM2 demonstrates superior performance across the evaluations. The predictive performance of AMP activity cliffs remains to be further improved, considering that ESM2 with 33 layers only achieves the Spearman correlation coefficient 0.4669 for the regression task of the MIC values on the benchmark dataset. Source code and additional resources are available at https://www.healthinformaticslab.org/supp/ or https://github.com/Kewei2023/AMPCliff-generation.

Yinheng Li, Han Ding, Shaofei Wang

Self-supervised learning has become a popular approach in recent years for its ability to learn meaningful representations without the need for data annotation. This paper proposes a novel image augmentation technique, overlaying images, which has not been widely applied in self-supervised learning. This method is designed to provide better guidance for the model to understand underlying information, resulting in more useful representations. The proposed method is evaluated using contrastive learning, a widely used self-supervised learning method that has shown solid performance in downstream tasks. The results demonstrate the effectiveness of the proposed augmentation technique in improving the performance of self-supervised models.

Hong Lu, Chuan Li, Yinheng Li et al.

The drug development process necessitates that pharmacologists undertake various tasks, such as reviewing literature, formulating hypotheses, designing experiments, and interpreting results. Each stage requires accessing and querying vast amounts of information. In this abstract, we introduce a Large Language Model (LLM)-based Natural Language Interface designed to interact with structured information stored in databases. Our experiments demonstrate the feasibility and effectiveness of the proposed framework. This framework can generalize to query a wide range of pharmaceutical data and knowledge bases.

Lawrence Jang, Yinheng Li, Dan Zhao et al.

Videos are often used to learn or extract the necessary information to complete tasks in ways different than what text and static imagery alone can provide. However, many existing agent benchmarks neglect long-context video understanding, instead focusing on text or static image inputs. To bridge this gap, we introduce VideoWebArena (VideoWA), a benchmark for evaluating the capabilities of long-context multimodal agents for video understanding. VideoWA consists of 2,021 web agent tasks based on manually crafted video tutorials, which total almost four hours of content. For our benchmark, we define a taxonomy of long-context video-based agent tasks with two main areas of focus: skill retention and factual retention. While skill retention tasks evaluate whether an agent can use a given human demonstration to complete a task efficiently, the factual retention task evaluates whether an agent can retrieve instruction-relevant information from a video to complete a task. We find that the best model achieves 13.3% success on factual retention tasks and 45.8% on factual retention QA pairs, far below human performance at 73.9% and 79.3%, respectively. On skill retention tasks, long-context models perform worse with tutorials than without, exhibiting a 5% performance decrease in WebArena tasks and a 10.3% decrease in VisualWebArena tasks. Our work highlights the need to improve the agentic abilities of long-context multimodal models and provides a testbed for future development with long-context video agents.

Tianyi Chen, Yinheng Li, Michael Solodko et al.

Computer-Using Agents (CUAs) aim to autonomously operate computer systems to complete real-world tasks. However, existing agentic systems remain difficult to scale and lag behind human performance. A key limitation is the absence of reusable and structured skill abstractions that capture how humans interact with graphical user interfaces and how to leverage these skills. We introduce CUA-Skill, a computer-using agentic skill base that encodes human computer-use knowledge as skills coupled with parameterized execution and composition graphs. CUA-Skill is a large-scale library of carefully engineered skills spanning common Windows applications, serving as a practical infrastructure and tool substrate for scalable, reliable agent development. Built upon this skill base, we construct CUA-Skill Agent, an end-to-end computer-using agent that supports dynamic skill retrieval, argument instantiation, and memory-aware failure recovery. Our results demonstrate that CUA-Skill substantially improves execution success rates and robustness on challenging end-to-end agent benchmarks, establishing a strong foundation for future computer-using agent development. On WindowsAgentArena, CUA-Skill Agent achieves state-of-the-art 57.5% (best of three) successful rate while being significantly more efficient than prior and concurrent approaches. The project page is available at https://microsoft.github.io/cua_skill/.

Saeed Amizadeh, Sara Abdali, Yinheng Li et al.

Transformers and their attention mechanism have been revolutionary in the field of Machine Learning. While originally proposed for the language data, they quickly found their way to the image, video, graph, etc. data modalities with various signal geometries. Despite this versatility, generalizing the attention mechanism to scenarios where data is presented at different scales from potentially different modalities is not straightforward. The attempts to incorporate hierarchy and multi-modality within transformers are largely based on ad hoc heuristics, which are not seamlessly generalizable to similar problems with potentially different structures. To address this problem, in this paper, we take a fundamentally different approach: we first propose a mathematical construct to represent multi-modal, multi-scale data. We then mathematically derive the neural attention mechanics for the proposed construct from the first principle of entropy minimization. We show that the derived formulation is optimal in the sense of being the closest to the standard Softmax attention while incorporating the inductive biases originating from the hierarchical/geometric information of the problem. We further propose an efficient algorithm based on dynamic programming to compute our derived attention mechanism. By incorporating it within transformers, we show that the proposed hierarchical attention mechanism not only can be employed to train transformer models in hierarchical/multi-modal settings from scratch, but it can also be used to inject hierarchical information into classical, pre-trained transformer models post training, resulting in more efficient models in zero-shot manner.

Yinheng Li, Hailey Hultquist, Justin Wagle et al.

Graphical user interface (GUI) agents have advanced rapidly but still struggle with complex tasks involving novel UI elements, long-horizon actions, and personalized trajectories. In this work, we introduce Instruction Agent, a GUI agent that leverages expert demonstrations to solve such tasks, enabling completion of otherwise difficult workflows. Given a single demonstration, the agent extracts step-by-step instructions and executes them by strictly following the trajectory intended by the user, which avoids making mistakes during execution. The agent leverages the verifier and backtracker modules further to improve robustness. Both modules are critical to understand the current outcome from each action and handle unexpected interruptions(such as pop-up windows) during execution. Our experiments show that Instruction Agent achieves a 60% success rate on a set of tasks in OSWorld that all top-ranked agents failed to complete. The Instruction Agent offers a practical and extensible framework, bridging the gap between current GUI agents and reliable real-world GUI task automation.

Yinheng Li, Rogerio Bonatti, Sara Abdali et al.

Using Large Language Models (LLMs) to generate synthetic data for model training has become increasingly popular in recent years. While LLMs are capable of producing realistic training data, the effectiveness of data generation is influenced by various factors, including the choice of prompt, task complexity, and the quality, quantity, and diversity of the generated data. In this work, we focus exclusively on using synthetic data for text classification tasks. Specifically, we use natural language understanding (NLU) models trained on synthetic data to assess the quality of synthetic data from different generation approaches. This work provides an empirical analysis of the impact of these factors and offers recommendations for better data generation practices.

Sander Schulhoff, Michael Ilie, Nishant Balepur et al.

Generative Artificial Intelligence (GenAI) systems are increasingly being deployed across diverse industries and research domains. Developers and end-users interact with these systems through the use of prompting and prompt engineering. Although prompt engineering is a widely adopted and extensively researched area, it suffers from conflicting terminology and a fragmented ontological understanding of what constitutes an effective prompt due to its relatively recent emergence. We establish a structured understanding of prompt engineering by assembling a taxonomy of prompting techniques and analyzing their applications. We present a detailed vocabulary of 33 vocabulary terms, a taxonomy of 58 LLM prompting techniques, and 40 techniques for other modalities. Additionally, we provide best practices and guidelines for prompt engineering, including advice for prompting state-of-the-art (SOTA) LLMs such as ChatGPT. We further present a meta-analysis of the entire literature on natural language prefix-prompting. As a culmination of these efforts, this paper presents the most comprehensive survey on prompt engineering to date.

Yinheng Li, Junhao Wang, Yijie Cao

Portfolio management is the art and science in fiance that concerns continuous reallocation of funds and assets across financial instruments to meet the desired returns to risk profile. Deep reinforcement learning (RL) has gained increasing interest in portfolio management, where RL agents are trained base on financial data to optimize the asset reallocation process. Though there are prior efforts in trying to combine RL and portfolio management, previous works did not consider practical aspects such as transaction costs or short selling restrictions, limiting their applicability. To address these limitations, we propose a general RL framework for asset management that enables continuous asset weights, short selling and making decisions with relevant features. We compare the performance of three different RL algorithms: Policy Gradient with Actor-Critic (PGAC), Proximal Policy Optimization (PPO), and Evolution Strategies (ES) and demonstrate their advantages in a simulated environment with transaction costs. Our work aims to provide more options for utilizing RL frameworks in real-life asset management scenarios and can benefit further research in financial applications.