RoboHub's platform is a central hub for robotics enthusiasts, offering insights into robotics research, robot design, and autonomous systems development. Through articles, interviews, and research papers, RoboHub offers insights into robotic technologies, artificial intelligence algorithms, and human-robot interaction. Readers can learn about robot sensing and perception, motion planning, and control strategies to design and build innovative robotic systems for various applications.

Robohub

Aniket Roy, a recent Johns Hopkins PhD graduate now at NEC Laboratories America, discusses his research on resource-constrained image generation and visual understanding. His work spans few-shot learning (FeLMi), caption-guided augmentation (Cap2Aug), perceptual quality improvement in diffusion models via kurtosis concentration loss (DiffNat), and parameter-efficient personalization using LoRA-based frameworks (DuoLoRA, MultiLFG). A key insight is that classical image statistics — specifically kurtosis consistency across wavelet frequency bands — can improve perceptual realism in modern diffusion models without full retraining. His research targets making large generative models more efficient and adaptable under data and compute constraints.

Resource-constrained image generation and visual understanding: an interview with Aniket Roy