Scholar
Daniil Dmitriev
Google Scholar ID: 3_4gF8wAAAAJ
ETH Zurich
Citations & Impact
Citations
348
H-index
4
i10-index
3
Publications
13
Co-authors
17
list available
Publications
Resume (English only)
Academic Achievements
- Published 'Robust mixture learning when outliers overwhelm small groups' at NeurIPS 2024, proposing an efficient meta-algorithm for recovering mixture means under large additive contamination.
- Published 'On the growth of mistakes in differentially private online learning: a lower bound perspective' at COLT 2024, proving logarithmic mistake growth under differential privacy constraints.
- Published 'Asymptotics of learning with deep structured (random) features' at ICML 2024, deriving deterministic equivalents for generalization error and proving linearization of sample covariance matrices in two-layer structured random feature models.
- Published 'Deterministic equivalent and error universality of deep random features learning' at ICML 2023, rigorously establishing Gaussian universality for test error in ridge regression with frozen intermediate layers.
- Published 'The Lovász number of random circulant graphs' at SampTA 2025, providing upper and lower bounds for the expected Lovász theta number of random circulant graphs.
- Published 'Greedy heuristics and linear relaxations for the random hitting set problem' at APPROX 2024, showing the standard greedy algorithm is order-optimal for the random Bernoulli hitting set problem.
- Published 'On monotonicity of Ramanujan function for binomial random variables' in Statistics & Probability Letters 2021, analyzing CDF properties of binomial variables near the median.
- Published 'Dynamic model pruning with feedback' at ICLR 2020, proposing a novel model compression method that yields sparse trained models without extra overhead.
- Invited talks include DACO seminar at ETH Zurich (2024), 'Youth in High Dimensions' at ICTP Trieste (2024), Delta seminar at University of Copenhagen (2024), and Graduate seminar in probability at ETH Zurich.
