İlker Oğuz

Wavefront-Shaping Enabled Scalable Optical Circuit Switch

Niyazi Ulaş Dinç, Mustafa Yıldırım, İlker Oğuz, Demetri Psaltis et autres

Optical interconnects accelerate AI training, yet Optical Circuit Switching (OCS) adoption lags due to reliability, scalability, and flexibility challenges. We propose leveraging wave optics with distributed light modulation, enabling CMOS-compatible, low-voltage, nonblocking switching. This versatile approach allows miniaturization and easier scaling, overcoming …

ch (code pays fourni par la source)

Exploring Expert Specialization through Unsupervised Training in Sparse Mixture of Experts

İlker Oğuz, Demetri Psaltis

Understanding the internal organization of neural networks remains a fundamental challenge in deep learning interpretability. We address this challenge by exploring a novel Sparse Mixture of Experts Variational Autoencoder (SMoE-VAE) architecture. We test our model on the QuickDraw dataset, comparing unsupervised expert …

Training of physical neural networks

Ali Momeni, Babak Rahmani, Benjamin Scellier, Logan G. Wright et autres

ch, gb, us, de, fr, jp, it (code pays fourni par la source)

Multilayer Light Processor for AI Compute and Networking

Mustafa Yıldırım, Niyazi Ulaş Dinç, İlker Oğuz, Demetri Psaltis et autres

This paper investigates optical processors for AI, addressing interconnect bottlenecks and optical data processing. We introduce a Multilayer Light Processor, a programmable 3D free-space optical architecture for neural processing and optical circuit switching.

gb, ch (code pays fourni par la source)

Training hybrid neural networks with multimode optical nonlinearities using digital twins

İlker Oğuz, Louis J. E. Suter, Jih-Liang Hsieh, Mustafa Yildirim et autres

The ability to train ever-larger neural networks brings artificial intelligence to the forefront of scientific and technical discoveries. However, their exponentially increasing size creates a proportionally greater demand for energy and computational hardware. Incorporating complex physical events in networks as fixed, efficient …

ch (code pays fourni par la source)

Optical diffusion models for image generation

Mustafa Yıldırım, İlker Oğuz, Niyazi Ulaş Dinç, Junjie Ke et autres

Using AI-driven wavefront-shaping for scalable optical circuit switching (Conference Presentation)

Niyazi Ulaş Dinç, Mustafa Yıldırım, İlker Oğuz, Christophe Moser et autres

The demand for efficient, scalable, and high-capacity interconnects has surged in the rapidly evolving landscape of data center networks. While the current optical interconnect technologies offer benefits over electrical interconnects, they fall short in the scalability of the port count in a …

ch (code pays fourni par la source)

Roadmap on Neuromorphic Photonics

Daniel Brunner, Bhavin J. Shastri, Mohammed Qadasi, Hitesh Ballani et autres

This roadmap consolidates recent advances while exploring emerging applications, reflecting the remarkable diversity of hardware platforms, neuromorphic concepts, and implementation philosophies reported in the field. It emphasizes the critical role of cross-disciplinary collaboration in this rapidly evolving field.

fr, be, de, us, ca (code pays fourni par la source)

Training Hybrid Neural Networks with Multimode Optical Nonlinearities Using Digital Twins

İlker Oğuz, Louis J. E. Suter, Jih-Liang Hsieh, Mustafa Yıldırım et autres

The ability to train ever-larger neural networks brings artificial intelligence to the forefront of scientific and technical discoveries. However, their exponentially increasing size creates a proportionally greater demand for energy and computational hardware. Incorporating complex physical events in networks as fixed, efficient …

Resource-efficient photonic networks for next-generation AI computing

İlker Oğuz, Mustafa Yıldırım, Jih-Liang Hsieh, Niyazi Ulaş Dinç et autres

Current trends in artificial intelligence toward larger models demand a rethinking of both hardware and algorithms. Photonics-based systems offer high-speed, energy-efficient computing units, provided algorithms are designed to exploit photonics' unique strengths. The recent implementation of cellular automata in photonics demonstrates how …

ch (code pays fourni par la source)

On the implementation of nonlinearities in optical neural networks: opinion

Niyazi Ulaş Dinç, İlker Oğuz, Mustafa Yıldırım, Christophe Moser et autres

Optical neural networks are expected to be hybrid systems for the foreseeable future combining optics and electronics. The linear portion of the system, consisting of the weighted interconnections between the neurons, can be realized optically. The nonlinear activation functions and the training …

Nonlinear processing with linear optics

Mustafa Yildirim, Niyazi Ulaş Dinç, İlker Oğuz, Demetri Psaltis et autres

Deep neural networks have achieved remarkable breakthroughs by leveraging multiple layers of data processing to extract hidden representations, albeit at the cost of large electronic computing power. To enhance energy efficiency and speed, the optical implementation of neural networks aims to harness …

ch (code pays fourni par la source)