arXiv subjects

Zixi Zhang

Publications and source records attributed to Zixi Zhang.

8 recordsLinked to original sources

arXiv2026

Team of Thoughts: Efficient Test-time Scaling of Agentic Systems through Orchestrated Tool Calling

Existing Multi-Agent Systems (MAS) typically rely on homogeneous model configurations, failing to exploit the diverse expertise inherent in different post-trained architectures. We propose Team-of-Thoughts, a heterogeneous MAS framework that treats diverse models as specialized tools within an orchestrator-driven paradigm. Team-of-Thoughts introduces two novel components: (1) Orchestrator Calibration, which identifies models with superior coordination and synthesis capabilities, and (2) Agent Self-Assessment, a protocol where tool agents profile their own domain-specific strengths to guide selection. At inference, the orchestrator dynamically activates the most compatible agents based on these profiles to maximize capability coverage. Across five mathematical reasoning and code generation benchmarks, Team-of-Thoughts consistently outperforms individual models and existing MAS baselines. Notably, on AIME24 and LiveCodeBench, Team-of-Thoughts achieves 96.00% and 77.91% accuracy, respectively, significantly improving over homogeneous role-play baselines (80.00% and 65.93%).

cs.CL↗

arXiv2026

Deep Kernel Fusion for Transformers

Agentic LLM inference with long contexts is increasingly limited by memory bandwidth rather than compute. In this setting, SwiGLU MLP blocks, whose large weights exceed cache capacity, become a major yet under-optimized bottleneck. We propose DeepFusionKernel, a deeply fused kernel that cuts HBM traffic and boosts cache reuse, delivering up to 13.2% speedup on H100 and 9.7% on A100 over SGLang. Integrated with SGLang and paired with a kernel scheduler, DeepFusionKernel ensures consistent accelerations over generation lengths, while remaining adaptable to diverse models, inference configurations, and hardware platforms.

cs.LG↗

arXiv2025

Group Policy Gradient

We introduce Group Policy Gradient (GPG), a family of critic-free policy-gradient estimators for general MDPs. Inspired by the success of GRPO's approach in Reinforcement Learning from Human Feedback (RLHF), GPG replaces a learned value function with a group-based Monte Carlo advantage estimator, removing the memory, compute, and hyperparameter costs of training a critic while preserving PPO's clipped-objective structure. We prove the consistency of the GPG estimator, analyze the bias-variance tradeoffs, and demonstrate empirically that GPG matches or outperforms PPO on standard benchmarks. GPG makes better use of parallel simulations, which, together with its critic-free design, results in more efficient use of computational resources than PPO.

cs.LG↗

arXiv2025

Ab initio study of flexoelectricity in MXene monolayers

We investigate flexoelectricity in MXene monolayers from first principles. Specifically, we compute the transverse flexoelectric coefficients of 126 MXene monolayers along their two principal directions using Kohn-Sham density functional theory. The values span a wide range from 0.19$e$ to 1.3$e$ and are nearly isotropic with respect to bending direction. The transition metal is found to play a significant role in the flexoelectric response, with nitride-based MXenes consistently displaying larger coefficients than their carbide counterparts. Moreover, the coefficients increase with structural thickness, but when normalized by the bending modulus, which is also computed for all 126 monolayers, they exhibit the opposite trend.

cond-mat.mtrl-sci↗

arXiv2025

LLM4DV: Using Large Language Models for Hardware Test Stimuli Generation

Hardware design verification (DV) is a process that checks the functional equivalence of a hardware design against its specifications, improving hardware reliability and robustness. A key task in the DV process is the test stimuli generation, which creates a set of conditions or inputs for testing. These test conditions are often complex and specific to the given hardware design, requiring substantial human engineering effort to optimize. We seek a solution of automated and efficient testing for arbitrary hardware designs that takes advantage of large language models (LLMs). LLMs have already shown promising results for improving hardware design automation, but remain under-explored for hardware DV. In this paper, we propose an open-source benchmarking framework named LLM4DV that efficiently orchestrates LLMs for automated hardware test stimuli generation. Our analysis evaluates six different LLMs involving six prompting improvements over eight hardware designs and provides insight for future work on LLMs development for efficient automated DV.

cs.LG↗

arXiv2024

Unlocking the Global Synergies in Low-Rank Adapters

Low-rank Adaption (LoRA) has been the de-facto parameter-efficient fine-tuning technique for large language models. We present HeteroLoRA, a light-weight search algorithm that leverages zero-cost proxies to allocate the limited LoRA trainable parameters across the model for better fine-tuned performance. In addition to the allocation for the standard LoRA-adapted models, we also demonstrate the efficacy of HeteroLoRA by performing the allocation in a more challenging search space that includes LoRA modules and LoRA-adapted shortcut connections. Experiments show that HeteroLoRA enables improvements in model performance given the same parameter budge. For example, on MRPC, we see an improvement of 1.6% in accuracy with similar training parameter budget. We will open-source our algorithm once the paper is accepted.

cs.LG↗

arXiv2023

Electromechanical memcapacitive neurons for energy-efficient spiking neural networks

In this article, we introduce a new nanoscale electromechanical device -- a leaky memcapacitor -- and show that it may be useful for the hardware implementation of spiking neurons. The leaky memcapacitor is a movable-plate capacitor that becomes quite conductive when the plates come close to each other. The equivalent circuit of the leaky memcapacitor involves a memcapacitive and memristive system connected in parallel. In the leaky memcapacitor, the resistance and capacitance depend on the same internal state variable, which is the displacement of the movable plate. We have performed a comprehensive analysis showing that several spiking types observed in biological neurons can be implemented with the leaky memcapacitor. Significant attention is paid to the dynamic properties of the model. As in leaky memcapacitors the capacitive and leaking resistive functionalities are implemented naturally within the same device structure, their use will simplify the creation of spiking neural networks.

cs.ET↗

arXiv2023

Graphic characterization and clustering configuration descriptors of determinant space for molecules

Quantum Monte Carlo approaches based on the stochastic sampling of the determinant space have evolved to be powerful methods to compute the electronic states of molecules. These methods not only calculate the correlation energy at an unprecedented accuracy but also provides insightful information on the electronic structure of computed states, e.g. the population, connection, and clustering of determinants, which have not been fully explored. In this work, we devise a configuration graph for visualizing the determinant space, revealing the nature of the molecule's electronic structure. In addition, we propose two analytical descriptors to quantify the extent of configuration clustering of multi-determinant wave functions. The graph and descriptors provide us with a fresh perspective of the electronic structure of molecules and can assist the further development of configuration interaction based electronic structure methods.

physics.chem-ph↗