IBM · Agora

Selected work

Representative Papers

On-line Policy Improvement using Monte-Carlo Search

Dec 03, 1996Neural Information Processing Systems

This paper addresses the low decision quality and high error rates of controllers in real-time adaptive control. We propose an online Monte Carlo Policy Improvement (MCPI) algorithm that requires neither an environmental model nor gradient information, relying solely on a simulatable environment and an initial policy. MCPI estimates long-term action returns via parallel multi-step random rollouts and dynamically updates the policy. Its key innovation lies in directly applying a lightweight, scalable Monte Carlo Tree Search (MCTS) for online policy optimization, enabling plug-and-play reinforcement learning enhancement. Evaluated on backgammon, MCPI reduces decision error rates by over fivefold compared to baselines—including random policies and TD-Gammon—demonstrating strong generalization capability and real-time efficacy in practical adaptive control scenarios.

268 citations20 influentialRead paper

A 3D particle visualization system for temperature management

Jan 23, 2011Electronic imaging

This paper addresses the challenge of real-time visualization and energy-efficiency analysis for massive sensor data (temperature, humidity, pressure) in data centers. To this end, we propose an adaptive particle-based modeling method tailored for thermal-field sensing. Built upon a client-server architecture, the approach integrates an enhanced Level-of-Detail (LOD) multi-scale simplification strategy—extending Clark’s 1976 theory—and pioneers the synergistic use of particle systems with dynamic detail control for data center thermal management. The system achieves millisecond-level, three-dimensional streaming rendering of tens of thousands of sensor streams, improving frame rate by 3.2×. It significantly enhances hotspot localization accuracy and accelerates energy-efficiency regulation response. As a result, it establishes a scalable, real-time thermal-state visualization paradigm for green data centers.

22 citations1 influentialRead paper

Application of quantum machine learning using quantum kernel algorithms on multiclass neuron M-type classification

Jul 17, 2023Scientific Reports

This study investigates the applicability of quantum machine learning to multi-class neuronal M-type classification—a key challenge in computational neuroscience and electrophysiological signal analysis. We propose the first quantum kernel method specifically designed for multi-class morphological neuronal classification, and develop a scalable quantum–classical hybrid feature mapping framework integrating parameterized quantum circuits, quantum kernel embedding, and classical support vector machines (SVMs). Experiments are conducted on both synthetic and real neuronal electrophysiological datasets, with model training performed jointly on the Qiskit simulator and IBM Quantum hardware. Results demonstrate an average classification accuracy of 92.4%, outperforming classical SVM by 6.8 percentage points. To our knowledge, this is the first empirical validation of quantum kernel methods achieving superior performance—and suggestive quantum advantage—in multi-class discriminative tasks within neuroscience.

11 citationsRead paper

Non-Abelian qLDPC: TQFT Formalism, Addressable Gauging Measurement and Application to Magic State Fountain on 2D Product Codes

Jan 11, 2026

This work addresses the challenge of reconciling connectivity and universality in two-dimensional architectures for fault-tolerant quantum computation with qLDPC codes. By generalizing Kitaev’s non-Abelian topological code to non-Abelian qLDPC codes, the authors construct a combinatorial topological quantum field theory based on Poincaré CW complexes and introduce a spacetime path integral formulation to enable addressable gauge measurements. The key innovation lies in the first realization of native non-Clifford logical gates on constant-rate two-dimensional hypergraph product codes, achieved through an addressable measurement scheme rooted in 0-form subcomplex symmetries, which is further extended to higher-dimensional and higher-order symmetries. This approach is successfully applied to magic state distillation, enabling the parallel preparation of $O(\sqrt{n})$ disjoint CZ magic states, each with code distance $O(\sqrt{n})$, on $n$ physical qubits.

4 citations2 influentialRead paper

Memorization to Generalization: Emergence of Diffusion Models from Associative Memory

May 27, 2025

This work investigates the memory–generalization phase transition in diffusion models under varying training data scales. We propose a *correlational memory* perspective: training corresponds to memory encoding, while generation implements memory retrieval. We establish, for the first time, a theoretical connection between diffusion models and Hopfield networks, deriving necessary and sufficient conditions for the emergence of *spurious attractors*—hallucinated states—at the critical memory load threshold. Leveraging energy landscape analysis, dynamical systems modeling, and empirical validation on DDPM and DDIM, we confirm the universality of this phenomenon. Results show that models operate dominantly in memory mode under small-data regimes, shift toward generalization with large-scale data, and exhibit spurious attractors in the critical regime—unifying explanations for memory overload and implicit manifold learning. This work provides a cross-disciplinary theoretical framework and falsifiable predictions for understanding the intrinsic mechanisms of diffusion models.

3 citationsRead paper

Recent publications