Selected work

Representative Papers

The Global Neural World Model: Spatially Grounded Discrete Topologies for Action-Conditioned Planning

Apr 17, 2026

This work addresses the challenges of manifold drift and inadequate spatial structure representation in autoregressive prediction for continuous environment modeling. To overcome these issues, the authors propose a self-stabilizing, action-conditioned Joint Embedding Predictive Architecture (JEPA) that maps the environment onto a discrete two-dimensional grid. By incorporating translation-equivariant constraints and a grid “snapping” mechanism, the model achieves structured world representation. A topological quantization strategy based on balanced continuous entropy regularization is introduced to embed error correction directly into the latent space, thereby mitigating manifold drift. Furthermore, maximum-entropy exploration encourages learning of generalizable dynamics rather than memorizing trajectories. Experiments demonstrate that the model functions effectively as both a spatial physics simulator and a causal discovery system across passive observation, active control, and abstract sequential tasks, successfully constructing structured topological maps.

0 citationsRead paper

Modular Continual Learning via Zero-Leakage Reconstruction Routing and Autonomous Task Discovery

Apr 15, 2026

This work addresses the dual challenges of catastrophic forgetting and data privacy in sequential learning with artificial neural networks by proposing a silicon-native modular architecture. The approach enforces parameter isolation through task-specific expert modules and an outlier-based distributed gating mechanism, while leveraging tight-bottleneck autoencoders to establish strict topological boundaries in high-dimensional embedding spaces, enabling unsupervised novelty detection and stable replay of known manifolds. During training, teacher learning, student distillation, and routing manifold acquisition are executed in parallel, and raw input data is discarded immediately after use, guaranteeing zero privacy leakage. Experiments demonstrate that the framework effectively mitigates forgetting in both computer vision and natural language processing tasks, preserving strong memory retention without compromising student fidelity, while fully complying with privacy regulations such as GDPR.

0 citationsRead paper

AI-assisted Human-in-the-Loop Web Platform for Structural Characterization in Hard drive design

Mar 31, 2026

This study addresses the limitations of rigid automated pipelines and inefficient, subjective manual approaches in analyzing semiconductor multilayer thin films from scanning transmission electron microscopy (STEM) images. To overcome these challenges, the authors propose a tunable human–AI collaborative metrology workflow that integrates human prior knowledge with algorithmic automation. The modular framework incorporates gradient peak detection, noise suppression, interface tracking, and interactive correction algorithms, enabling human intervention during design while supporting fully automatic execution during runtime. Operating directly on TEM/EMD file formats, the system delivers nanometer-precision measurements of layer thickness and interface roughness with statistical rigor. By balancing flexibility with automation, the approach enhances both analytical efficiency and consistency. The implementation is open-source, and the architecture is designed for reusability and extensibility.

0 citationsRead paper

From Bits to Chips: An LLM-based Hardware-Aware Quantization Agent for Streamlined Deployment of LLMs

Jan 07, 2026arXiv.org

Deploying large language models on resource-constrained hardware faces significant challenges, including the trade-off between accuracy and efficiency and the complexity of quantization hyperparameter tuning. This work proposes the Hardware-Aware Quantization Agent (HAQA), which leverages a large language model to automatically optimize quantization hyperparameters and adapt them to target hardware, enabling cross-platform adaptive quantization strategies. By doing so, HAQA substantially reduces the need for manual intervention and streamlines the deployment pipeline. Experiments on the Llama family of models demonstrate that HAQA achieves up to 2.3× speedup in inference latency and throughput while maintaining or even improving model accuracy, outperforming conventional non-optimized deployment approaches.

0 citationsRead paper

Forget-It-All: Multi-Concept Machine Unlearning via Concept-Aware Neuron Masking

Jan 07, 2026arXiv.org

This work addresses the challenge of efficiently and effectively unlearning multiple concepts—such as copyrighted or sensitive content—from text-to-image diffusion models without retraining. The authors propose a plug-and-play, training-free framework that identifies neurons associated with target concepts through contrastive concept saliency analysis and constructs a unified multi-concept mask by integrating spatial and temporal information for precise neuron pruning. By employing a neuron mask fusion strategy, the method achieves low hyperparameter sensitivity and eliminates the need for additional training, significantly improving unlearning performance across three benchmark forgetting tasks while preserving the semantic fidelity and visual quality of generated images.

0 citationsRead paper

Recent publications

Latest Papers

The Global Neural World Model: Spatially Grounded Discrete Topologies for Action-Conditioned Planning

Apr 17, 2026

This work addresses the challenges of manifold drift and inadequate spatial structure representation in autoregressive prediction for continuous environment modeling. To overcome these issues, the authors propose a self-stabilizing, action-conditioned Joint Embedding Predictive Architecture (JEPA) that maps the environment onto a discrete two-dimensional grid. By incorporating translation-equivariant constraints and a grid “snapping” mechanism, the model achieves structured world representation. A topological quantization strategy based on balanced continuous entropy regularization is introduced to embed error correction directly into the latent space, thereby mitigating manifold drift. Furthermore, maximum-entropy exploration encourages learning of generalizable dynamics rather than memorizing trajectories. Experiments demonstrate that the model functions effectively as both a spatial physics simulator and a causal discovery system across passive observation, active control, and abstract sequential tasks, successfully constructing structured topological maps.

0 citationsRead paper

Modular Continual Learning via Zero-Leakage Reconstruction Routing and Autonomous Task Discovery

Apr 15, 2026

This work addresses the dual challenges of catastrophic forgetting and data privacy in sequential learning with artificial neural networks by proposing a silicon-native modular architecture. The approach enforces parameter isolation through task-specific expert modules and an outlier-based distributed gating mechanism, while leveraging tight-bottleneck autoencoders to establish strict topological boundaries in high-dimensional embedding spaces, enabling unsupervised novelty detection and stable replay of known manifolds. During training, teacher learning, student distillation, and routing manifold acquisition are executed in parallel, and raw input data is discarded immediately after use, guaranteeing zero privacy leakage. Experiments demonstrate that the framework effectively mitigates forgetting in both computer vision and natural language processing tasks, preserving strong memory retention without compromising student fidelity, while fully complying with privacy regulations such as GDPR.

0 citationsRead paper

AI-assisted Human-in-the-Loop Web Platform for Structural Characterization in Hard drive design

Mar 31, 2026

This study addresses the limitations of rigid automated pipelines and inefficient, subjective manual approaches in analyzing semiconductor multilayer thin films from scanning transmission electron microscopy (STEM) images. To overcome these challenges, the authors propose a tunable human–AI collaborative metrology workflow that integrates human prior knowledge with algorithmic automation. The modular framework incorporates gradient peak detection, noise suppression, interface tracking, and interactive correction algorithms, enabling human intervention during design while supporting fully automatic execution during runtime. Operating directly on TEM/EMD file formats, the system delivers nanometer-precision measurements of layer thickness and interface roughness with statistical rigor. By balancing flexibility with automation, the approach enhances both analytical efficiency and consistency. The implementation is open-source, and the architecture is designed for reusability and extensibility.

0 citationsRead paper

From Bits to Chips: An LLM-based Hardware-Aware Quantization Agent for Streamlined Deployment of LLMs

Jan 07, 2026arXiv.org

Deploying large language models on resource-constrained hardware faces significant challenges, including the trade-off between accuracy and efficiency and the complexity of quantization hyperparameter tuning. This work proposes the Hardware-Aware Quantization Agent (HAQA), which leverages a large language model to automatically optimize quantization hyperparameters and adapt them to target hardware, enabling cross-platform adaptive quantization strategies. By doing so, HAQA substantially reduces the need for manual intervention and streamlines the deployment pipeline. Experiments on the Llama family of models demonstrate that HAQA achieves up to 2.3× speedup in inference latency and throughput while maintaining or even improving model accuracy, outperforming conventional non-optimized deployment approaches.

0 citationsRead paper

Forget-It-All: Multi-Concept Machine Unlearning via Concept-Aware Neuron Masking

Jan 07, 2026arXiv.org

This work addresses the challenge of efficiently and effectively unlearning multiple concepts—such as copyrighted or sensitive content—from text-to-image diffusion models without retraining. The authors propose a plug-and-play, training-free framework that identifies neurons associated with target concepts through contrastive concept saliency analysis and constructs a unified multi-concept mask by integrating spatial and temporal information for precise neuron pruning. By employing a neuron mask fusion strategy, the method achieves low hyperparameter sensitivity and eliminates the need for additional training, significantly improving unlearning performance across three benchmark forgetting tasks while preserving the semantic fidelity and visual quality of generated images.

0 citationsRead paper