Sattam Altuuaim

Sattam Altuuaim, Lama Ayash, Muhammad Mubashar et al.

Despite the central role of optimization in deep learning, most optimizers rely on update structures whose functional form is fixed before training begins. This static design can limit their ability to respond to changing gradient behavior across the loss landscape, where training may shift between stable, noisy, and inconsistent regimes. This study proposes PILOT (Policy-Informed Learned OpTimizer), an online optimizer that adapts its update behavior during training. Rather than using a fixed balance between momentum, normalization, and sign-based updates, PILOT uses gradient-direction agreement as a signal of local training stability. Conditioning the update rule on this agreement signal allows the optimizer to adjust its behavior when gradients become stable, noisy, or inconsistent. Experiments on FashionMNIST and CIFAR-10 show that PILOT consistently achieves the highest accuracy among the evaluated optimizers across convolutional settings. On the CNN architecture, PILOT reaches 94.13% on FashionMNIST and 81.94% on CIFAR-10. On ResNet-18, it further improves performance, reaching 95.71% on FashionMNIST and 93.42% on CIFAR-10. These results suggest that learning how to adapt the update structure during training can improve performance across both compact and deeper convolutional models while preserving a simple first-order optimization framework. The implementation of PILOT is publicly available at https://github.com/SattamAltwaim/PILOT.git

Eyad Alghamdi, Sattam Altuuaim, Obay Ghulam et al.

Arabic Sign Language (ArSL) and its dialects serve approximately 400 million Arabic speakers worldwide, yet the community lacks high-quality 3D parametric annotations and specialized reconstruction methods for avatar generation. We address this critical gap through two key contributions: First, we introduce the first high-quality 3D parametric annotations for the Ishara-500 Saudi Sign Language dataset, providing precise SMPL-X parameters for 500 culturally authentic SSL signs. Second, we present Tamaththul3D, a specialized reconstruction pipeline designed for ArSL's unique articulation patterns. Our pipeline integrates SMPLer-X for robust body estimation, WiLoR for detailed hand refinement with automatic localization and mirroring, and MediaPipe for 2D pose supervision. Through kinematic-chain-based wrist alignment with hybrid swing-twist decomposition and 2D-supervised joint optimization, Tamaththul3D achieves state-of-the-art hand accuracy (up to 32% improvement over previous methods) while maintaining competitive body pose. Together, these 3D annotations and Tamaththul3D pipeline establish the first comprehensive framework for high-fidelity ArSL avatar reconstruction, enabling new accessibility technologies and cultural preservation efforts for the Arab Deaf community.