معرفی
Jona Ballé is an Associate Professor in the Electrical and Computer Engineering department at New York University's Tandon School of Engineering. His research focuses on developing efficient representations of visual media through machine learning and end-to-end optimization techniques.
Dr. Ballé's research interests center on visual media compression, spanning still images, video, augmented reality, virtual reality, plenoptic imaging, and holographic imaging. His work bridges information theory, computer vision, and machine learning to develop perceptually optimized compression algorithms. He has made significant contributions to understanding the relationship between human visual perception and image statistics, which has led to improved compression results and ultimately contributed to the JPEG AI standard finalized in 2025.
His recent publications demonstrate a strong trend toward Wasserstein distortion metrics, neural compression architectures, and rate-distortion optimization. These works span computer vision, information theory, and signal processing domains, with applications in both traditional and emerging visual media formats. His research shows consistent innovation in developing perceptually relevant metrics that balance fidelity and realism in compressed media.
- Contributed to JPEG AI standard (2025)
- Co-organizer of Challenge on Learned Image Compression (CLIC) since 2018
- Program committee member of Data Compression Conference (DCC) since 2022
- Reviewer for top-tier publications including NeurIPS, ICLR, ICML, and IEEE Transactions journals
Dr. Ballé has advised numerous graduate students who have co-authored significant publications with him, particularly in the areas of neural compression and perceptual metrics. His research has been supported by institutions including the Simons Foundation. He maintains active collaborations across academia and industry, with his work at Google (2017-2024) directly informing his current academic research.
His laboratory focuses on developing open-source implementations of advanced compression techniques, with notable GitHub repositories including Wasserstein Distortion implementation in PyTorch and CoDeX (Learned data compression in JAX), demonstrating his commitment to reproducible research and community engagement.
