Lincen Bai

Lincen Bai, Hedi Tabia, Raul Santos-Rodriguez

Pruning vision-language models (VLMs) for efficient deployment is challenging because compression can affect not only task utility but also visual grounding, often amplifying object hallucinations even at the same sparsity level. We present HiPP-Prune, a hierarchical preference-conditioned structured pruning framework that treats pruning as conditional resource allocation under multiple objectives. HiPP-Prune makes plan-level decisions: a single policy invocation outputs a global pruning blueprint by factorizing decisions into an overall sparsity budget and a layer-wise allocation, enabling queryable trade-offs via a user-specified preference vector. To account for VLM-specific failure modes, our policy state integrates a visual sensitivity signal derived from attention flow between vision tokens and language hidden states, discouraging over-pruning of vision-critical layers that facilitate cross-modal fusion. We optimize pruning plans with plan-level Group Relative Policy Optimization (GRPO) under a multi-objective return that combines task utility, hallucination robustness (POPE), compression, and a synaptic-flow-inspired stability proxy to reduce unproductive exploration in high-sparsity regimes. Experiments on LLaVA with POPE and ScienceQA demonstrate that HiPP-Prune discovers diverse non-dominated pruning plans and provides controllable robustness--utility trade-offs under matched sparsity budgets.

Hongyang Li, Lincen Bai, Caesar Wu et al.

We propose LQ-SGD (Low-Rank Quantized Stochastic Gradient Descent), an efficient communication gradient compression algorithm designed for distributed training. LQ-SGD further develops on the basis of PowerSGD by incorporating the low-rank approximation and log-quantization techniques, which drastically reduce the communication overhead, while still ensuring the convergence speed of training and model accuracy. In addition, LQ-SGD and other compression-based methods show stronger resistance to gradient inversion than traditional SGD, providing a more robust and efficient optimization path for distributed learning systems.

Lincen Bai, Hedi Tabia, Raúl Santos-Rodríguez

Neural network pruning has traditionally focused on weight-based criteria to achieve model compression, frequently overlooking the crucial balance between adversarial robustness and accuracy. Existing approaches often fail to preserve robustness in pruned networks, leaving them more susceptible to adversarial attacks. This paper introduces Module Robustness Sensitivity (MRS), a novel metric that quantifies layer-wise sensitivity to adversarial perturbations and dynamically informs pruning decisions. Leveraging MRS, we propose Module Robust Pruning and Fine-Tuning (MRPF), an adaptive pruning algorithm compatible with any adversarial training method, offering both flexibility and scalability. Extensive experiments on SVHN, CIFAR, and Tiny-ImageNet across diverse architectures, including ResNet, VGG, and MobileViT, demonstrate that MRPF significantly enhances adversarial robustness while maintaining competitive accuracy and computational efficiency. Furthermore, MRPF consistently outperforms state-of-the-art structured pruning methods in balancing robustness, accuracy, and compression. This work establishes a practical and generalizable framework for robust pruning, addressing the long-standing trade-off between model compression and robustness preservation.