Publications

TinyFoA: Memory Efficient Forward-Only Algorithm for On-Device Learning

In this paper, we propose a memory-efficient forward-only algorithm called TinyFoA, to reduce dynamic memory overhead in the training process. Our TinyFoA optimizes the memory efficiency not only by layer-wise training but also by partially updating each layer, as well as by binarizing the weights and the activations.

AAAI Conference on Artificial Intelligence (AAAI) (2025)

EEG Glasses for Real-Time Brain Electrical Activity Monitoring

In this work, we present e-Glass, a state-of-the-art smart wearable device that enables unobtrusive real-time electroencephalography (EEG) monitoring. Moreover, we present a lightweight edge-ML scheme tailored for e-Glass, which has limited resources (compute, memory, and energy).

Scientific Reports (2025)

VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees

In this paper, we introduce a post-training framework to create Verification-Friendly Neural Networks (VNNs) that are easier for formal verification tools to analyze, allowing for provable robustness guarantees, while maintaining accuracy comparable to standard Deep Neural Networks (DNNs).

International Conference on Machine Learning (ICML) (2024)

Privacy-Preserving Edge Federated Learning for Intelligent Mobile-Health Systems

In this work, we propose a privacy-preserving edge federated learning framework for resource-constrained mobile-health over the Internet of Things (IoT) infrastructure. Our framework guarantees that the partial contribution of each patient during the learning process remains private, by adopting Secure Multiparty Computation (SMC) techniques.

Future Generation Computer Systems (2024)