Hamza ElMokhtar Shili

Yehya Farhat, Hamza ElMokhtar Shili, Fangshuo Liao et al.

Mixture-of-Experts (MoEs) achieve scalability by dynamically activating subsets of their components. Yet, understanding how expertise emerges through joint training of gating mechanisms and experts remains incomplete, especially in scenarios without clear task partitions. Motivated by inference costs and data heterogeneity, we study how joint training of gating functions and experts can dynamically allocate domain-specific expertise across multiple underlying data distributions. As an outcome of our framework, we develop an instance tailored specifically to decentralized training scenarios, introducing \textit{Dynamically Decentralized Orchestration of MoEs} or \texttt{DDOME}. \texttt{DDOME} leverages heterogeneity emerging from distributional shifts across decentralized data sources to specialize experts dynamically. By integrating a pretrained common expert to inform a gating function, \texttt{DDOME} achieves personalized expert subset selection on-the-fly, facilitating just-in-time personalization. We empirically validate \texttt{DDOME} within a Federated Learning (FL) context: \texttt{DDOME} attains from 4\% up to an 24\% accuracy improvement over state-of-the-art FL baselines in image and text classification tasks, while maintaining competitive zero-shot generalization capabilities. Furthermore, we provide theoretical insights confirming that the joint gating-experts training is critical for achieving meaningful expert specialization.

Hamza ElMokhtar Shili, Natasha Patnaik, Isabelle Ruble et al.

We investigate algorithmic variants of the Frank-Wolfe (FW) optimization method for pruning convolutional neural networks. This is motivated by the "Lottery Ticket Hypothesis", which suggests the existence of smaller sub-networks within larger pre-trained networks that perform comparatively well (if not better). Whilst most literature in this area focuses on Deep Neural Networks more generally, we specifically consider Convolutional Neural Networks for image classification tasks. Building on the hypothesis, we compare simple magnitude-based pruning, a Frank-Wolfe style pruning scheme, and an FW method with momentum on a CNN trained on MNIST. Our experiments track test accuracy, loss, sparsity, and inference time as we vary the dense pre-training budget from 1 to 10 epochs. We find that FW with momentum yields pruned networks that are both sparser and more accurate than the original dense model and the simple pruning baselines, while incurring minimal inference-time overhead in our implementation. Moreover, FW with momentum reaches these accuracies after only a few epochs of pre-training, indicating that full pre-training of the dense model is not required in this setting.