Anand Kumar

Currently, I am a Machine Learning Research Scientist at Collinear AI, where I lead the development of cutting-edge generative AI solutions. My work focuses on advancing the capabilities of large language and vision models (LLMs and VLMs) and diffusion models, particularly in the context of post-training adaptation and fine-tuning.

Previously, I was a Computer Vision researcher at UC San Diego’s Statistical Visual Computing Lab (SVCL), where I work under Prof. Nuno’s guidance to push the boundaries of generative AI and 3D reconstruction. My research focuses on developing innovative approaches using diffusion models, particularly in style attribution and sparse-view 3D reconstruction. Through my work on IntroStyle and GS-TransUNet, I’ve demonstrated a knack for developing novel solutions that advance the state-of-the-art in computer vision.

My academic journey spans from NIT Tiruchirappalli to UC San Diego’s Master’s program in Electrical and Computer Engineering, where I’ve cultivated expertise in computer vision, GPU programming, and statistical learning. This strong foundation in both theory and practice has enabled me to tackle complex challenges across various domains, from implementing CUDA-optimized optical flow algorithms to developing zero-shot image captioning frameworks.

Beyond academic achievements, I bring hands-on experience from diverse research environments, including remote work with Omnyk Inc. on sleep stage prediction and an internship at Leibniz University Hannover's TNT Lab, where I worked on video compression using VAE-GANs. These experiences have honed my ability to bridge theoretical concepts with practical implementations, particularly in deep learning frameworks like PyTorch, TensorFlow, and JAX. My current research at SVCL focuses on developing training-free approaches for style attribution, where our methods have shown significant improvements over existing benchmarks.