The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Deep Learning in Optics with Aydogan Ozcan - TWiML Talk #237
Mar 7, 2019
Aydogan Ozcan, a UCLA professor specializing in electrical and computer engineering, discusses pioneering research at the intersection of deep learning and optics. He explains the concept of all-optical neural networks that mimic neuron behavior through diffraction. The conversation dives into practical applications such as enhanced biomedical imaging and environmental monitoring. Ozcan also highlights the role of 3D printing in reducing costs and dimensions of optical networks, showcasing the possibilities for future innovations in defense and security.
AI Snips
Chapters
Transcript
Episode notes
Deep Learning for Hologram Reconstruction
- Deep learning reconstructs holograms faster than traditional physics-based iterative methods.
- It also offers better rejection of interference artifacts common in holography.
Holography's Information Channels
- Holography captures both amplitude and phase information of a sample, unlike traditional lens-based microscopy.
- Traditional holographic reconstruction methods serve as better ground truth labels for training deep learning models.
Autofocusing with Holography
- Holographic reconstruction with deep learning can reconstruct objects at different depths without physical focusing.
- Training with holograms from varied distances extends the depth of field, enabling digital autofocusing.