arXiv subjects

Minghao Li

Publications and source records attributed to Minghao Li.

At least 55 records · Page 3Linked to original sources

arXiv2025

Fusion Intelligence for Digital Twinning AI Data Centers: A Synergistic GenAI-PhyAI Approach

The explosion in artificial intelligence (AI) applications is pushing the development of AI-dedicated data centers (AIDCs), creating management challenges that traditional methods and standalone AI solutions struggle to address. While digital twins are beneficial for AI-based design validation and operational optimization, current AI methods for their creation face limitations. Specifically, physical AI (PhyAI) aims to capture the underlying physical laws, which demands extensive, case-specific customization, and generative AI (GenAI) can produce inaccurate or hallucinated results. We propose Fusion Intelligence, a novel framework synergizing GenAI's automation with PhyAI's domain grounding. In this dual-agent collaboration, GenAI interprets natural language prompts to generate tokenized AIDC digital twins. Subsequently, PhyAI optimizes these generated twins by enforcing physical constraints and assimilating real-time data. Case studies demonstrate the advantages of our framework in automating the creation and validation of AIDC digital twins. These twins deliver predictive analytics to support power usage effectiveness (PUE) optimization in the design stage. With operational data collected, the digital twin accuracy is further improved compared with pure physics-based models developed by human experts. Fusion Intelligence offers a promising pathway to accelerate digital transformation. It enables more reliable and efficient AI-driven digital transformation for a broad range of mission-critical infrastructures.

cs.AI↗

arXiv2025

PHYBench: Holistic Evaluation of Physical Perception and Reasoning in Large Language Models

Current benchmarks for evaluating the reasoning capabilities of Large Language Models (LLMs) face significant limitations: task oversimplification, data contamination, and flawed evaluation items. These deficiencies necessitate more rigorous assessment methods. To address these limitations, we introduce PHYBench, a benchmark of 500 original physics problems ranging from high school to Physics Olympiad difficulty. PHYBench addresses data contamination through original content and employs a systematic curation pipeline to eliminate flawed items. Evaluations show that PHYBench activates more tokens and provides stronger differentiation between reasoning models compared to other baselines like AIME 2024, OlympiadBench and GPQA. Even the best-performing model, Gemini 2.5 Pro, achieves only 36.9% accuracy compared to human experts' 61.9%. To further enhance evaluation precision, we introduce the Expression Edit Distance (EED) Score for mathematical expression assessment, which improves sample efficiency by 204% over binary scoring. Moreover, PHYBench effectively elicits multi-step and multi-condition reasoning, providing a platform for examining models' reasoning robustness, preferences, and deficiencies. The benchmark results and dataset are publicly available at https://www.phybench.cn/.

cs.CL↗

arXiv2025

Transforming Future Data Center Operations and Management via Physical AI

Data centers (DCs) as mission-critical infrastructures are pivotal in powering the growth of artificial intelligence (AI) and the digital economy. The evolution from Internet DC to AI DC has introduced new challenges in operating and managing data centers for improved business resilience and reduced total cost of ownership. As a result, new paradigms, beyond the traditional approaches based on best practices, must be in order for future data centers. In this research, we propose and develop a novel Physical AI (PhyAI) framework for advancing DC operations and management. Our system leverages the emerging capabilities of state-of-the-art industrial products and our in-house research and development. Specifically, it presents three core modules, namely: 1) an industry-grade in-house simulation engine to simulate DC operations in a highly accurate manner, 2) an AI engine built upon NVIDIA PhysicsNemo for the training and evaluation of physics-informed machine learning (PIML) models, and 3) a digital twin platform built upon NVIDIA Omniverse for our proposed 5-tier digital twin framework. This system presents a scalable and adaptable solution to digitalize, optimize, and automate future data center operations and management, by enabling real-time digital twins for future data centers. To illustrate its effectiveness, we present a compelling case study on building a surrogate model for predicting the thermal and airflow profiles of a large-scale DC in a real-time manner. Our results demonstrate its superior performance over traditional time-consuming Computational Fluid Dynamics/Heat Transfer (CFD/HT) simulation, with a median absolute temperature prediction error of 0.18 °C. This emerging approach would open doors to several potential research directions for advancing Physical AI in future DC operations.

cs.AI↗

arXiv2025

On the Classification of Isoparametric Hypersurfaces with Constant Principal Curvatures in Compact 3-Manifolds

Establishing detailed relationships between transnormal systems of different types and their behaviors under covering maps, this paper presents a classification of transnormal systems on compact 3-manifolds in the sense of equivalence. For CPC transnormal systems, we show that the ambient manifolds must be locally isometric to one of six standard geometries up to equivalence. We also find some equivalence classes containing no CPC transnormal system, highlighting a critical distinction between isoparametric foliations and CPC transnormal systems, which has not been previously addressed in the literature.

math.DG↗

arXiv2025

Towards Robust and Secure Embodied AI: A Survey on Vulnerabilities and Attacks

Embodied AI systems, including robots and autonomous vehicles, are increasingly integrated into real-world applications, where they encounter a range of vulnerabilities stemming from both environmental and system-level factors. These vulnerabilities manifest through sensor spoofing, adversarial attacks, and failures in task and motion planning, posing significant challenges to robustness and safety. Despite the growing body of research, existing reviews rarely focus specifically on the unique safety and security challenges of embodied AI systems. Most prior work either addresses general AI vulnerabilities or focuses on isolated aspects, lacking a dedicated and unified framework tailored to embodied AI. This survey fills this critical gap by: (1) categorizing vulnerabilities specific to embodied AI into exogenous (e.g., physical attacks, cybersecurity threats) and endogenous (e.g., sensor failures, software flaws) origins; (2) systematically analyzing adversarial attack paradigms unique to embodied AI, with a focus on their impact on perception, decision-making, and embodied interaction; (3) investigating attack vectors targeting large vision-language models (LVLMs) and large language models (LLMs) within embodied systems, such as jailbreak attacks and instruction misinterpretation; (4) evaluating robustness challenges in algorithms for embodied perception, decision-making, and task planning; and (5) proposing targeted strategies to enhance the safety and reliability of embodied AI systems. By integrating these dimensions, we provide a comprehensive framework for understanding the interplay between vulnerabilities and safety in embodied AI.

cs.CR↗

arXiv2025

The uniqueness of minimal maps into Cartan-Hadamard manifolds via the squared singular values

In this paper, we give a uniqueness theorem for the Dirichlet problem of minimal maps into general Riemannian manifolds with non-positive sectional curvature, improving Theorem 5.2 of Lee-Ooi-Tsui's paper published in J. Geom. Anal.. The proof of this theorem is based on the convexity of several functions in terms of squared singular values along the geodesic homotopy of two given minimal maps.

math.DG↗

arXiv2025

Some Remarks on Isoparametric Functions in Closed 4-Manifolds

We study transnormal and isoparametric functions on closed Riemannian 4-manifolds and establish fundamental restrictions on their topology and geometry. In particular, we show that such manifolds cannot be endowed with negatively curved metrics, contrasting with known results in the compact simply connected case. Moreover, in certain cases, we provide a description of their fundamental groups. These findings contribute to a better understanding of the global structure of isoparametric foliations.

math.GT↗

arXiv2025

The Equivalence of the Existences of Transnormal and Isoparametric Functions on Compact Manifolds

Through exploring the embedded transnormal systems of codimension 1, we show the existence of a transnormal function on a connected complete Riemannian manifold requires the underlying manifold to have a vector bundle structure or a linear double disk bundle decomposition. Conversely, any smooth manifold with either of these structures can be endowed with a Riemannian metric so that it admits a transnormal function, which, under suitable compactness conditions, can become isoparametric. As a corollary, for compact manifolds, the existences of transnormal and isoparametric functions impose the same topological constraints.

math.DG↗

arXiv2024

Adaptive Optimizers with Sparse Group Lasso for Neural Networks in CTR Prediction

We develop a novel framework that adds the regularizers of the sparse group lasso to a family of adaptive optimizers in deep learning, such as Momentum, Adagrad, Adam, AMSGrad, AdaHessian, and create a new class of optimizers, which are named Group Momentum, Group Adagrad, Group Adam, Group AMSGrad and Group AdaHessian, etc., accordingly. We establish theoretically proven convergence guarantees in the stochastic convex settings, based on primal-dual methods. We evaluate the regularized effect of our new optimizers on three large-scale real-world ad click datasets with state-of-the-art deep learning models. The experimental results reveal that compared with the original optimizers with the post-processing procedure which uses the magnitude pruning method, the performance of the models can be significantly improved on the same sparsity level. Furthermore, in comparison to the cases without magnitude pruning, our methods can achieve extremely high sparsity with significantly better or highly competitive performance. The code is available at https://github.com/intelligent-machine-learning/tfplus/tree/main/tfplus.

cs.LG↗

arXiv2024

Taming a Maxwell's demon for experimental stochastic resetting

A diffusive process that is reset to its origin at random times, so-called stochastic resetting (SR), is an ubiquitous expedient in many natural systems . Yet, beyond its ability to improve the efficiency of target searching, SR is a true non-equilibrium thermodynamic process that brings forward new and challenging questions . Here, we show how the recent developments of experimental information thermodynamics renew the way to address SR and can lead, beyond a new understanding, to better control on the non-equilibrium nature of SR. This thermodynamically controlled SR is experimentally implemented within a time-dependent optical trapping potential. We show in particular that SR converts heat into work from a single bath continuously and without feedback. This implements a Maxwell's demon that constantly erases information. In our experiments, the erasure takes the form of a protocol that allows to evaluate the true energetic cost of SR. We show that using an appropriate measure of the available information, this cost can be reduced to a reversible minimum while being bounded by the Landauer limit. We finally reveal that the individual trajectories generated by the demon all break ergodicity and thus demonstrate the non-ergodic nature of the demon's modus operandi. Our results offer new approaches to processes, such as SR, where the informational framework provides key experimental tools for their non-equilibrium thermodynamic control.

cond-mat.stat-mech↗

arXiv2024

Low-lying magnon frequency comb in skymion crystals

A stable, low-power and tunable magnon frequency comb (MFC) is crucial for magnon-based precision measurements, quantum information processing and chip integration. Original method for creating MFC utilizes the nonlinear interactions between propagating spin waves and localized oscillations of an isolated magnetic texture, e.g., skyrmion. It requires a driving frequency well above the ferromagnetic resonance (FMR) and the spectrum frequency of MFC will quickly approach to the detection limit of conventional microwave technique after only tens of comb teeth. In addition, the detection and manipulation of a single skyrmion is challenging in experiments due to its high degree of locality. These issues hinder the applications of MFC. In this work, we report the low-lying MFC with comb frequencies below the FMR in a skyrmion crystal (SkX). We show that the MFC originates from the three-wave mixing between the collective skyrmion gyration and breathing in the SkX. Our findings significantly improve the efficiency of the nonlinear frequency conversion from a single-frequency mircowave input, and establish a synergistic relationship between the SkX and MFC, which paves the way to coherent information processing and ultra-sensitive metrology based on MFC.

cond-mat.mtrl-sci↗

arXiv2024

UniCorn: A Unified Contrastive Learning Approach for Multi-view Molecular Representation Learning

Recently, a noticeable trend has emerged in developing pre-trained foundation models in the domains of CV and NLP. However, for molecular pre-training, there lacks a universal model capable of effectively applying to various categories of molecular tasks, since existing prevalent pre-training methods exhibit effectiveness for specific types of downstream tasks. Furthermore, the lack of profound understanding of existing pre-training methods, including 2D graph masking, 2D-3D contrastive learning, and 3D denoising, hampers the advancement of molecular foundation models. In this work, we provide a unified comprehension of existing pre-training methods through the lens of contrastive learning. Thus their distinctions lie in clustering different views of molecules, which is shown beneficial to specific downstream tasks. To achieve a complete and general-purpose molecular representation, we propose a novel pre-training framework, named UniCorn, that inherits the merits of the three methods, depicting molecular views in three different levels. SOTA performance across quantum, physicochemical, and biological tasks, along with comprehensive ablation study, validate the universality and effectiveness of UniCorn.

q-bio.BM↗

arXiv2024

THC: Accelerating Distributed Deep Learning Using Tensor Homomorphic Compression

Deep neural networks (DNNs) are the de facto standard for essential use cases, such as image classification, computer vision, and natural language processing. As DNNs and datasets get larger, they require distributed training on increasingly larger clusters. A main bottleneck is the resulting communication overhead where workers exchange model updates (i.e., gradients) on a per-round basis. To address this bottleneck and accelerate training, a widely-deployed approach is compression. However, previous deployments often apply bi-directional compression schemes by simply using a uni-directional gradient compression scheme in each direction. This results in significant computational overheads at the parameter server and increased compression error, leading to longer training and lower accuracy. We introduce Tensor Homomorphic Compression (THC), a novel bi-directional compression framework that enables the direct aggregation of compressed values and thus eliminating the aforementioned computational overheads. Moreover, THC is compatible with in-network aggregation (INA), which allows for further acceleration. Our evaluation shows that training representative vision and language models with THC reaches target accuracy by 1.40x to 1.47x faster using INA and 1.28x to 1.33x faster using a software PS compared with state-of-the-art systems.

cs.LG↗

arXiv2024

Preference Ranking Optimization for Human Alignment

Large language models (LLMs) often contain misleading content, emphasizing the need to align them with human values to ensure secure AI systems. Reinforcement learning from human feedback (RLHF) has been employed to achieve this alignment. However, it encompasses two main drawbacks: (1) RLHF exhibits complexity, instability, and sensitivity to hyperparameters in contrast to SFT. (2) Despite massive trial-and-error, multiple sampling is reduced to pair-wise contrast, thus lacking contrasts from a macro perspective. In this paper, we propose Preference Ranking Optimization (PRO) as an efficient SFT algorithm to directly fine-tune LLMs for human alignment. PRO extends the pair-wise contrast to accommodate preference rankings of any length. By iteratively contrasting candidates, PRO instructs the LLM to prioritize the best response while progressively ranking the rest responses. In this manner, PRO effectively transforms human alignment into aligning the probability ranking of n responses generated by LLM with the preference ranking of humans towards these responses. Experiments have shown that PRO outperforms baseline algorithms, achieving comparable results to ChatGPT and human responses through automatic-based, reward-based, GPT-4, and human evaluations.

cs.CL↗

arXiv2023

A note on the uniqueness of minimal maps into $\mathbb{R}^n$ via singular values

In this note, we derive a uniqueness theorem for minimal graphs of general codimension under certain restrictions closed related to the convexity (not strict convexity) of the area functional with respect to singular values, improving the result in \cite{L-O-T}. The crucial step of the proof is to show the local linearity of the singular value vectors along the geodesic homotopy of two given minimal maps.

math.DG↗

arXiv2023

Protein-ligand binding representation learning from fine-grained interactions

The binding between proteins and ligands plays a crucial role in the realm of drug discovery. Previous deep learning approaches have shown promising results over traditional computationally intensive methods, but resulting in poor generalization due to limited supervised data. In this paper, we propose to learn protein-ligand binding representation in a self-supervised learning manner. Different from existing pre-training approaches which treat proteins and ligands individually, we emphasize to discern the intricate binding patterns from fine-grained interactions. Specifically, this self-supervised learning problem is formulated as a prediction of the conclusive binding complex structure given a pocket and ligand with a Transformer based interaction module, which naturally emulates the binding process. To ensure the representation of rich binding information, we introduce two pre-training tasks, i.e.~atomic pairwise distance map prediction and mask ligand reconstruction, which comprehensively model the fine-grained interactions from both structure and feature space. Extensive experiments have demonstrated the superiority of our method across various binding tasks, including protein-ligand affinity prediction, virtual screening and protein-ligand docking.

q-bio.BM↗

arXiv2023

API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs

Recent research has demonstrated that Large Language Models (LLMs) can enhance their capabilities by utilizing external tools. However, three pivotal questions remain unanswered: (1) How effective are current LLMs in utilizing tools? (2) How can we enhance LLMs' ability to utilize tools? (3) What obstacles need to be overcome to leverage tools? To address these questions, we introduce API-Bank, a groundbreaking benchmark, specifically designed for tool-augmented LLMs. For the first question, we develop a runnable evaluation system consisting of 73 API tools. We annotate 314 tool-use dialogues with 753 API calls to assess the existing LLMs' capabilities in planning, retrieving, and calling APIs. For the second question, we construct a comprehensive training set containing 1,888 tool-use dialogues from 2,138 APIs spanning 1,000 distinct domains. Using this dataset, we train Lynx, a tool-augmented LLM initialized from Alpaca. Experimental results demonstrate that GPT-3.5 exhibits improved tool utilization compared to GPT-3, while GPT-4 excels in planning. However, there is still significant potential for further improvement. Moreover, Lynx surpasses Alpaca's tool utilization performance by more than 26 pts and approaches the effectiveness of GPT-3.5. Through error analysis, we highlight the key challenges for future research in this field to answer the third question.

cs.CL↗

arXiv2023

Functional estimation in high-dimensional and infinite-dimensional models

Let ${\mathcal P}$ be a family of probability measures on a measurable space $(S,{\mathcal A}).$ Given a Banach space $E,$ a functional $f:E\mapsto {\mathbb R}$ and a mapping $θ: {\mathcal P}\mapsto E,$ our goal is to estimate $f(θ(P))$ based on i.i.d. observations $X_1,\dots, X_n\sim P, P\in {\mathcal P}.$ In particular, if ${\mathcal P}=\{P_θ: θ\in Θ\}$ is an identifiable statistical model with parameter set $Θ\subset E,$ one can consider the mapping $θ(P)=θ$ for $P\in {\mathcal P}, P=P_θ,$ resulting in a problem of estimation of $f(θ)$ based on i.i.d. observations $X_1,\dots, X_n\sim P_θ, θ\in Θ.$ Given a smooth functional $f$ and estimators $\hat θ_n(X_1,\dots, X_n), n\geq 1$ of $θ(P),$ we use these estimators, the sample split and the Taylor expansion of $f(θ(P))$ of a proper order to construct estimators $T_f(X_1,\dots, X_n)$ of $f(θ(P)).$ For these estimators and for a functional $f$ of smoothness $s\geq 1,$ we prove upper bounds on the $L_p$-errors of estimator $T_f(X_1,\dots, X_n)$ under certain moment assumptions on the base estimators $\hat θ_n.$ We study the performance of estimators $T_f(X_1,\dots, X_n)$ in several concrete problems, showing their minimax optimality and asymptotic efficiency. In particular, this includes functional estimation in high-dimensional models with many low dimensional components, functional estimation in high-dimensional exponential families and estimation of functionals of covariance operators in infinite-dimensional subgaussian models.

math.ST↗