Rush Tabesh

Hey there, I’m Rush (Soroush) Tabesh, a Ph.D. student in Computer Science at ISTA. I’m lucky to work with Dan Alistarh in the DASLab, where we explore how to make deep learning leaner, faster, and easier to train and deploy.

Before ISTA, I earned my B.Sc. in Computer Science at Sharif University of Technology, where I worked with Mohammad Hossein Yassaee on knowledge distillation.

Research

I work on making neural networks cheaper to train and deploy—fewer bits, less memory, faster iterations. My current focus is on quantization-aware training (QAT) and fully quantized training, pushing models down to 4-bit, 3-bit, or even 1-bit precision while trying to preserve the accuracy. I’m also interested in parameter-efficient fine-tuning (PEFT) and neural network compression, finding ways to adapt or prune large models without the usual computational overhead.

If you’re curious to dive deeper into my research, feel free to check out my Google Scholar page for a full list of papers!

---

A Little Extra

When I’m not wrangling deep nets, I’m usually doing something completely unrelated, like slowly learning guitar (emphasis on slowly), going for short runs that feel longer than they should, or keep up with the thrilling process of learning German, which mostly means nodding politely and hoping context clues are on my side.