Alexandru Meterez — ameterez [at] g [dot] harvard [dot] edu

Hey, I'm Alex, and I am a CS PhD student at Harvard co-advised by Prof. Cengiz Pehlevan and Prof. Sham Kakade. I am fortunate to be supported by a Kempner Fellowship in my research. My main interests are in Deep Learning theory and neural network optimization.

I aim to understand how these models learn from data by developing a theoretical framework backed by empirical evidence. My research philosophy is to balance theoretical rigor with providing strong intuitions for observed phenomena.

I finished my Data Science MSc at ETH Zurich working with Gunnar Rätsch and Francesco Orabona (KAUST) and my CS BSc at UPB working with Iuliu Vasilescu. I also interned in various places.

Publications and preprints (*: equal co-first, †: equal co-last)

• (under review) Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining
  R. Zhao*, A. Meterez*, S. Kakade, C. Pehlevan, S. Jelassi†, E. Malach†
• [ICLR25] The Optimization Landscape of SGD Across the Feature Learning Strength
  A. Atanasov*, A. Meterez*, J. Simon*, C. Pehlevan
• [NeurIPS24] Why do Learning Rates Transfer? Reconciling Optimization and Scaling Limits for Deep Learning
  L. Noci*, A. Meterez*. T. Hofmann, A. Orvieto
• [ICLR24] Towards Training Without Depth Limits: Batch Normalization Without Gradient Explosion
  A. Meterez*, A. Joudaki*, F. Orabona, A. Immer, G. Rätsch, H. Daneshmand
• [RECOMB23] Aligning distant sequences to graphs using long seed sketches
  A. Joudaki*, A. Meterez*, H.Mustafa. R. Groot Koerkamp, A. Kahles, G. Rätsch

Misc.

I was born in Romania. I used to do a lot of systems programming back in my BSc years. I am also an advocate of open-source software. Some of my hobbies are running, reading, swimming, and since I mostly commute by bike, bike fixing.

parts of this website are borrowed from this work