Searcharxiv⌕ Search

arXiv subjects

Linliang Chen

Publications and source records attributed to Linliang Chen.

3 recordsLinked to original sources

GEAR$^2$: Pruned Experts Live On through Gated Reuse of Activated Experts and a Missing-Mass Residual

Mixture-of-Experts (MoE) Large Language Models (LLMs) activate only a few experts per token while generally keeping all the experts in GPU memory for high efficiency. Expert pruning reduces this GPU memory footprint cost, yet discards the contribution of pruned experts. Existing remedies either rely on static substitutes, merge experts at the cost of altering their functions, or require costly recovery training. In this paper, we find that the functionalities of pruned experts can be efficiently reconstructed from the activated remaining experts. We propose GEAR (Gated Expert Activation Reuse), which reuses the scaled outputs of reliable activated experts without any training or additional expert computation. In addition, we propose GEAR$^2$ to further recover what reuse cannot express with an ultra-lightweight residual. On five MoE LLMs from 16B to 106B parameters with 25% and 50% of experts pruned, GEAR improves over its training-free counterparts in nearly every setting (up to 4 pp.) in terms of accuracy, and GEAR$^2$ outperforms baseline approaches (up to 2.75 pp.) in every setting on nine zero-shot benchmarks, with larger advantages (up to 2.72 pp. higher) under more aggressive pruning (up to 50%).

cs.AI↗

BASC : Behavior-Aligned Quantization and Pruning for Low-Bit Spiking Neural Networks

Spiking Neural Networks (SNNs) encode information through binary spikes and compute in an event-driven manner, offering an energy-efficient paradigm for machine intelligence. However, high-performance SNNs incur substantial memory and timestep-wise computation costs that hinder deployment on resource-constrained devices. Quantization and pruning provide complementary routes to reducing these costs, yet both make their decisions with local criteria that overlook temporal task feedback in quantization and inter-channel dependencies in pruning. Consequently, optimizing either criterion can still yield suboptimal compression performance. We refer to this discrepancy as criterion-behavior mismatch and propose Behavior-Aligned SNN Compression (BASC), a unified framework with two lightweight modules. For quantization, the scale is applied to synaptic current at every timestep and therefore shifts spike timing. Temporal-Behavior Scale Correction (TSC) makes the scale learnable under a temporal loss, allowing firing behavior to inform scale optimization. For pruning, channel importance depends on how channels jointly drive the membrane potential across the firing threshold. Boundary-Level Inter-Channel Correction (BIC) uses channelwise importance scores for initial selection and inter-channel information to re-evaluate only channels near the pruning threshold. Extensive experiments on static and neuromorphic benchmarks show that lower-bit BASC models match or outperform higher-bit baselines and retain this accuracy advantage after structured pruning, while further reducing model storage and synaptic operations.

cs.NE↗

QUEST: A Quantized Energy-Aware SNN Training Framework for Multi-State Neuromorphic Devices

Neuromorphic devices, leveraging novel physical phenomena, offer a promising path toward energy-efficient hardware beyond CMOS technology by emulating brain-inspired computation. However, their progress is often limited to proof-of-concept studies due to the lack of flexible spiking neural network (SNN) algorithm frameworks tailored to device-specific characteristics, posing a significant challenge to scalability and practical deployment. To address this, we propose QUEST, a unified co-design framework that directly trains SNN for emerging devices featuring multilevel resistances. With Skyrmionic Magnetic Tunnel Junction (Sk-MTJ) as a case study, experimental results on the CIFAR-10 dataset demonstrate the framework's ability to enable scalable on-device SNN training with minimal energy consumption during both feedforward and backpropagation. By introducing device mapping pattern and activation operation sparsity, QUEST achieves effective trade-offs among high accuracy (89.6%), low bit precision (2-bit), and energy efficiency (93 times improvement over the ANNs). QUEST offers practical design guidelines for both the device and algorithm communities, providing insights to build energy-efficient and large-scale neuromorphic systems.

physics.app-ph↗