My final coffee with (some of) the Dirk Englund group members. Honoured to have these inspiring, hardworking and tenacious friends as colleagues. Gonna miss you all. From RLE at MIT Massachusetts Institute of Technology (MIT) -> Niels Bohr Institute in Copenhagen! Excited for the next adventure 💕

Splashdown of Crew Dragon in the Atlantic Ocean!

A great week for #PhotonicsWest.

Our fabulous Caterina Vigliar behind the lens today. Snappy art. #nofilter ^J

Here's the methodology we developed to perform design verification of the photonic, analog, and digital components of our processors: medium.com/lightmatter/te…

Jacob has been very inspiring and helpful to me. An incredibly well deserved award.

Our preprint on single cell multi-omic data integration w/ Gromov-Wasserstein optimal transport is up: biorxiv.org/content/10.110… This method uses fewer hyperparameters and is computationally more efficient compared to other unsupervised methods. All comments welcome!

biorxiv.org/content/10.110… #SingleCell #omics #optimalTransport #bioinformatics

Our work proposing optics as a means of creating freely scalable digital machine learning accelerators with no length-dependent cost is up on arXiv: arxiv.org/abs/2006.13926

`Freely scalable and reconfigurable optical hardware for deep learning" ArXiv arXiv:2006.13926 (2020) : scirate.com/arxiv/2006.139… It was vital to get at true bottlenecks by benchmarking studies with amazing electronics-based AI experts, Vivienne Sze and Prof Joel Emer - thanks!

Is it possible to program perfect optical transformations from imperfect components, without significant hardware overhead? Saumil, Ryan, and Dirk find the answer is yes through a deterministic algorithm, provided hardware errors are small enough: arxiv.org/abs/2103.04993

I'm very pleased to share my recent work experimentally demonstrating a photonic edge computing (cont) tl.gd/n_1srvrgp

Can high-performance machine learning run on low-power devices? After a few-second delay waiting for cloud-delegated inference, Apple's Siri might chime "No." But in a new article on ArXiv (arxiv.org/pdf/2203.05466…), we argue the opposite, ..

Dr Ryan Hamerly RLE at MIT NTT Research Xupi and Co. do a deep-dive on NetCast, our optical neural-network architecture designed for edge computing: arxiv.org/abs/2207.01777.

Now out in arXiv.org from MIT Prof. Dirk Englund, MIT Quantum Photonics, NTT Research & Xupi: "Netcast: Low-Power Edge Computing with WDM-defined Optical Neural Networks" by R. Hamerly, Alexander, S. Bandyopadhyay, Z. Chen, Z. Zhong, L. Bernstein, & Dirk Englund. arxiv.org/abs/2207.01777

In work by Dr Zaijun Chen et al , the Dirk Englund and Dr Ryan Hamerly AI team MIT Quantum Photonics RLE at MIT MIT EECS Microsystems Technology Laboratories, in collaboration with Prof Stephan Reitzenstein and Prof James A. Lott & groups of TU Berlin, introduce an ONN system based on a network of VCSEL-arrays..

Now in arXiv.org from MIT Prof. Dirk Englund, MIT Quantum Photonics, w/ TU Berlin: "Deep Learning with Coherent VCSEL Neural Networks" Zaijun Chen, Alexander, R. Davis, I. Christen, L. Bernstein, T. Heuser, N. Heermeier, J. A, Lott, S. Reitzenstein, R. Hamerly, D. Englund. arxiv.org/abs/2207.05329

New Science Magazine Podcast: Snakes living the high-altitude life w/ Elizabeth Pennisi, and sending computing power to the edges of the internet w/ Meagan Cantwell Alexander MIT Quantum Photonics science.org/doi/10.1126/sc…

Smart devices, such as cell phones and sensors, can't run the best machine learning models. The reason: large power and latency costs from moving model weights on size, weight and power constrained hardware. In this weeks issue of…lnkd.in/epSDw2T6 lnkd.in/eZ5uYupB

Check out the latest Science podcast with lead author Alexander, discussing network based deep learning : science.org/content/podcas…