Mikhail Belkin
Professor
Studies the theory of machine learning, including manifold regularization, interpolation and double descent in deep networks.
Mikhail Belkin's lab at UCSD works on learning theory & optimization and data management & big data. 26 papers since 2023, 23 at top venues, mostly ICLR and ICML. Applications close .
At a glance
- papers, 2023–2026
- 7/ year
- at T1 venues
- 88%
- PhD students
- —
- MS students
- —
- alumni listed
- —
Papers by yearfrom the lab homepage, 2023–2026
- 4 papers2026 · T1 3 · collecting
- 7 papers2025 · T1 7
- 7 papers2024 · T1 7
- 8 papers2023 · T1 6
T1 venuesOtherVenues: ICLR, ICML, NeurIPS, PNAS, Science
Research focuspaper titles that use each topic's words
- Learning theory & optimization5 papers
- Data management & big data1 paper
- deep networks0 papers
- manifold0 papers
- hal0 papers
Paper list
From the lab homepage.
20264 papers
- Applied and Computational Harmonic Analysis, 82, 101834Assembly and iteration: Transition to linearity of wide neural networks (opens in a new tab)
- ICLR (T1)The Features at Convergence Theorem: a first-principles alternative to the Neural Feature Ansatz for how networks learn representations (opens in a new tab)
- Science (T1)Toward universal steering and monitoring of AI models (opens in a new tab)
- ICLR (T1)xRFM: Accurate, scalable, and interpretable feature learning models for tabular data (opens in a new tab)
20257 papers
- COLT (T1)A Gap Between the Gaussian RKHS and Neural Networks: An Infinite-Center Asymptotic Analysis (opens in a new tab)
- ICML (T1)Emergence in non-neural models: grokking modular arithmetic via average gradient outer product (opens in a new tab)
- NeurIPS (T1)Fast training of large kernel models with delayed projections (opens in a new tab)
- PNAS (T1)Linear Recursive Feature Machines provably recover low-rank matrices (opens in a new tab)
- NeurIPS (T1)Seeds of Structure: Patch PCA Reveals Universal Compositional Cues in Diffusion Models (opens in a new tab)
- ICML (T1)Task Generalization with Autoregressive Compositional Structure: Can Learning from D Tasks Generalize to D^T Tasks? (opens in a new tab)
- NAACL (T1)Unmemorization in large language models via self-distillation and deliberate imagination (opens in a new tab)
20247 papers
- NeurIPS (T1)Average gradient outer product as a mechanism for deep neural collapse (opens in a new tab)
- ICML (T1)Catapults in SGD: spikes in the training loss and their impact on generalization through feature learning (opens in a new tab)
- Science (T1)Mechanism for feature learning in neural networks and backpropagation-free machine learning models (opens in a new tab)
- ICLR (T1)More is Better: when Infinite Overparameterization is Optimal and Overfitting is Obligatory (opens in a new tab)
- AISTATS (T1)On the Nyström Approximation for Preconditioning in Kernel Machines (opens in a new tab)
- ICLR (T1)Quadratic models for understanding catapult dynamics of neural networks (opens in a new tab)
- UAI (T1)Uncertainty Estimation with Recursive Feature Machines (opens in a new tab)
20238 papers
- SIAM Journal on Mathematics of Data ScienceA Universal Trade-off Between the Model Size, Test Loss, and Training Loss of Linear Predictors (opens in a new tab)
- NeurIPS (T1)Aiming towards the minimizers: fast convergence of SGD for overparametrized problems (opens in a new tab)
- ICML (T1)Cut your Losses with Squentropy (opens in a new tab)
- UAI (T1)Neural tangent kernel at initialization: linear width suffices (opens in a new tab)
- SIAM Journal on Mathematics of Data ScienceOn the Inconsistency of Kernel Ridgeless Regression in Fixed Dimensions (opens in a new tab)
- ICLR (T1)Restricted strong convexity of deep learning models with smooth activations (opens in a new tab)
- ICML (T1)Toward Large Kernel Models (opens in a new tab)
- PNAS (T1)Wide and deep neural networks achieve consistency for classification (opens in a new tab)
Where alumni went
Alumni are not listed on the lab homepage.
Apply to UCSDFall 2027
UCSD CSE Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- CNZhejiang
Deng Cai
- Recommendation, search & web
- CNTsinghua
Hairong Lv
- Trustworthy & safe ML
- GBImperial
Stefan Vlaski
- Learning theory & optimization
- CHEPFL
Lenka Zdeborová
- Learning theory & optimization
- SGNUS
Roland H. C. Yap
- Reasoning & neuro-symbolic AI
- USUCLA
Guy Van den Broeck
- Probabilistic & causal inference
Your notes
Save this lab to keep notes.
Data
Some numbers could not be verified:
- The member list could not be found on the lab site.
- The alumni list could not be found on the lab site.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Not collected
- Alumni
- Not collected
Tiers are venue classifications, not lab rankings.
UCSDcloses