arXiv subjects

Jeff Chen

Publications and source records attributed to Jeff Chen.

6 recordsLinked to original sources

arXiv2026

Cross-Domain Hybrid OPD for Generalizable Search Agents

Recent advances in Reinforcement Learning (RL) have substantially improved the capabilities of autonomous search agents, enabling sophisticated planning, and iterative retrieval over dynamic information sources. However, optimizing language models for specialized search behaviors often incurs an alignment tax, where gains in search performance come at the expense of general-purpose capabilities, limiting their effectiveness as universal assistants. In this technical report, we present the training framework behind the Yuanbao search agent, designed to achieve search specialization without sacrificing general intelligence. Built upon the Hunyuan3 architecture, our framework combines agentic reinforcement learning for autonomous search with a cross-domain expert On-Policy Distillation (OPD) pipeline. Experts specializing in complementary general-purpose domains are distilled into the search-specialized student, restoring and further enhancing its broad capabilities. Rather than treating specialization and general capability as competing objectives, our hybrid training strategy jointly optimizes both, effectively mitigating the alignment tax. Extensive experiments demonstrate that the resulting model achieves competitive search performance while consistently improving its general-purpose capabilities, providing a favorable balance between specialized execution and broad generalization in real-world search scenarios.

cs.CL↗

arXiv2025

Let the Barbarians In: How AI Can Accelerate Systems Performance Research

Artificial Intelligence (AI) is beginning to transform the research process by automating the discovery of new solutions. This shift depends on the availability of reliable verifiers, which AI-driven approaches require to validate candidate solutions. Research focused on improving systems performance is especially well-suited to this paradigm because system performance problems naturally admit such verifiers: candidates can be implemented in real systems or simulators and evaluated against predefined workloads. We term this iterative cycle of generation, evaluation, and refinement AI-Driven Research for Systems (ADRS). Using several open-source ADRS instances (i.e., OpenEvolve, GEPA, and ShinkaEvolve), we demonstrate across ten case studies (e.g., multi-region cloud scheduling, mixture-of-experts load balancing, LLM-based SQL, transaction scheduling) that ADRS-generated solutions can match or even outperform human state-of-the-art designs. Based on these findings, we outline best practices (e.g., level of prompt specification, amount of feedback, robust evaluation) for effectively using ADRS, and we discuss future research directions and their implications. Although we do not yet have a universal recipe for applying ADRS across all of systems research, we hope our preliminary findings, together with the challenges we identify, offer meaningful guidance for future work as researcher effort shifts increasingly toward problem formulation and strategic oversight. Note: This paper is an extension of our prior work [14]. It adds extensive evaluation across multiple ADRS frameworks and provides deeper analysis and insights into best practices.

cs.SE↗

arXiv2025

Barbarians at the Gate: How AI is Upending Systems Research

Artificial Intelligence (AI) is starting to transform the research process as we know it by automating the discovery of new solutions. Given a task, the typical AI-driven approach is (i) to generate a set of diverse solutions, and then (ii) to verify these solutions and select one that solves the problem. Crucially, this approach assumes the existence of a reliable verifier, i.e., one that can accurately determine whether a solution solves the given problem. We argue that systems research, long focused on designing and evaluating new performance-oriented algorithms, is particularly well-suited for AI-driven solution discovery. This is because system performance problems naturally admit reliable verifiers: solutions are typically implemented in real systems or simulators, and verification reduces to running these software artifacts against predefined workloads and measuring performance. We term this approach as AI-Driven Research for Systems (ADRS), which iteratively generates, evaluates, and refines solutions. Using penEvolve, an existing open-source ADRS instance, we present case studies across diverse domains, including load balancing for multi-region cloud scheduling, Mixture-of-Experts inference, LLM-based SQL queries, and transaction scheduling. In multiple instances, ADRS discovers algorithms that outperform state-of-the-art human designs (e.g., achieving up to 5.0x runtime improvements or 50% cost reductions). We distill best practices for guiding algorithm evolution, from prompt design to evaluator construction, for existing frameworks. We then discuss the broader implications for the systems community: as AI assumes a central role in algorithm design, we argue that human researchers will increasingly focus on problem formulation and strategic guidance. Our results highlight both the disruptive potential and the urgent need to adapt systems research practices in the age of AI.

cs.AI↗

arXiv2024

Tempus Core: Area-Power Efficient Temporal-Unary Convolution Core for Low-Precision Edge DLAs

The increasing complexity of deep neural networks (DNNs) poses significant challenges for edge inference deployment due to resource and power constraints of edge devices. Recent works on unary-based matrix multiplication hardware aim to leverage data sparsity and low-precision values to enhance hardware efficiency. However, the adoption and integration of such unary hardware into commercial deep learning accelerators (DLA) remain limited due to processing element (PE) array dataflow differences. This work presents Tempus Core, a convolution core with highly scalable unary-based PE array comprising of tub (temporal-unary-binary) multipliers that seamlessly integrates with the NVDLA (NVIDIA's open-source DLA for accelerating CNNs) while maintaining dataflow compliance and boosting hardware efficiency. Analysis across various datapath granularities shows that for INT8 precision in 45nm CMOS, Tempus Core's PE cell unit (PCU) yields 59.3% and 15.3% reductions in area and power consumption, respectively, over NVDLA's CMAC unit. Considering a 16x16 PE array in Tempus Core, area and power improves by 75% and 62%, respectively, while delivering 5x and 4x iso-area throughput improvements for INT8 and INT4 precisions. Post-place and route analysis of Tempus Core's PCU shows that the 16x4 PE array for INT4 precision in 45nm CMOS requires only 0.017 mm^2 die area and consumes only 6.2mW of total power. We demonstrate that area-power efficient unary-based hardware can be seamlessly integrated into conventional DLAs, paving the path for efficient unary hardware for edge AI inference.

cs.AR↗

arXiv2021

Multi-User Augmented Reality with Infrastructure-free Collaborative Localization

Multi-user augmented reality (AR) could someday empower first responders with the ability to see team members around corners and through walls. For this vision of people tracking in dynamic environments to be practical, we need a relative localization system that is nearly instantly available across wide-areas without any existing infrastructure or manual setup. In this paper, we present LocAR, an infrastructure-free 6-degrees-of-freedom (6DoF) localization system for AR applications that uses motion estimates and range measurements between users to establish an accurate relative coordinate system. We show that not only is it possible to perform collaborative localization without infrastructure or global coordinates, but that our approach provides nearly the same level of accuracy as fixed infrastructure approaches for AR teaming applications. LocAR uses visual-inertial odometry (VIO) in conjunction with ultra-wideband (UWB) ranging radios to estimate the relative position of each device in an ad-hoc manner. The system leverages a collaborative 6DoF particle filtering formulation that operates on sporadic messages exchanged between nearby users. Unlike map or landmark sharing approaches, this allows for collaborative AR sessions even if users do not overlap the same spaces. LocAR consists of an open-source UWB firmware and reference mobile phone application that can display the location of team members in real-time using mobile AR. We evaluate LocAR across multiple buildings under a wide-variety of conditions including a contiguous 30,000 square foot region spanning multiple floors and find that it achieves median geometric error in 3D of less than 1 meter between five users freely walking across 3 floors.

eess.SY↗

arXiv2021

Standing on the Shoulders of Machine Learning: Can We Improve Hypothesis Testing?

In this paper we have updated the hypothesis testing framework by drawing upon modern computational power and classification models from machine learning. We show that a simple classification algorithm such as a boosted decision stump can be used to fully recover the full size-power trade-off for any single test statistic. This recovery implies an equivalence, under certain conditions, between the basic building block of modern machine learning and hypothesis testing. Second, we show that more complex algorithms such as the random forest and gradient boosted machine can serve as mapping functions in place of the traditional null distribution. This allows for multiple test statistics and other information to be evaluated simultaneously and thus form a pseudo-composite hypothesis test. Moreover, we show how practitioners can make explicit the relative costs of Type I and Type II errors to contextualize the test into a specific decision framework. To illustrate this approach we revisit the case of testing for unit roots, a difficult problem in time series econometrics for which existing tests are known to exhibit low power. Using a simulation framework common to the literature we show that this approach can improve upon overall accuracy of the traditional unit root test(s) by seventeen percentage points, and the sensitivity by thirty six percentage points.

econ.EM↗