|
/ist/
|
IST
|
|
Анализировать url
|
|
/nonconvex-qst/
|
QST
|
|
Анализировать url
|
|
/explore-quantum/
|
QuanTAS
|
|
Анализировать url
|
|
/aiowls/
|
AI-OWLS
|
|
Анализировать url
|
|
/crysformer/
|
CrysFormer
|
|
Анализировать url
|
|
/group/
|
OptimaLab
|
|
Анализировать url
|
|
/pubs/
|
Publications
|
|
Анализировать url
|
|
/projects/
|
Software
|
|
Анализировать url
|
|
/about/
|
About me
|
|
Анализировать url
|
|
/2026/08/03/condition-number-tax-blog.html
|
<div class="padded-content">
<div class="title">The condition-number tax: same matrix, same condition number, three iterations or three thousand</div>
<!-- <div class="title">The condition-number tax: same matrix, same condition number, three iterations or three thousand</div> -->
<info datetime="">
Aug 2026
</info>
<span class="body">We published a new blog post on our quantum research page (QuanTAS), about the symbol that every quantum linear solver quotes its cost in — the condition number \(\kappa\) — and what it is actually measuring.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/07/15/maxkcut-theory-blog.html
|
<div class="padded-content">
<div class="title">The Sample-Complexity Theory Behind Randomized Max-3-Cut Rounding</div>
<!-- <div class="title">The Sample-Complexity Theory Behind Randomized Max-3-Cut Rounding</div> -->
<info datetime="">
Jul 2026
</info>
<span class="body">We published a new blog post on our explore-quantum page, working out the theory behind the randomized sampling algorithm used elsewhere in our low-rank Max-K-Cut series.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/07/08/which-layer-runs-the-program-blog.html
|
<div class="padded-content">
<div class="title">Which layer runs the program? What decides where a transformer computes</div>
<!-- <div class="title">Which layer runs the program? What decides where a transformer computes</div> -->
<info datetime="">
Jul 2026
</info>
<span class="body">We put up a new blog post on our AI-OWLS page, riding a question that is suddenly everywhere.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/07/07/can-we-train-a-computer-blog.html
|
<div class="padded-content">
<div class="title">Can we train a computer? Two ways to point at a memory</div>
<!-- <div class="title">Can we train a computer? Two ways to point at a memory</div> -->
<info datetime="">
Jul 2026
</info>
<span class="body">We put up a new blog post on our AI-OWLS page — the first of a short series of small, ground-truthed observations about how trained networks actually compute.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/07/04/ip-pdt-blog.html
|
<div class="padded-content">
<div class="title">When trainability outlasts expressibility: training quantum eigensolvers with far fewer gates</div>
<!-- <div class="title">When trainability outlasts expressibility: training quantum eigensolvers with far fewer gates</div> -->
<info datetime="">
Jul 2026
</info>
<span class="body">We published a new blog post on our quantum research page (QuanTAS), on a small structural trick that makes variational quantum eigensolvers easier to train — and a theorem that explains why it keeps helping long after it “should” have stopped.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/07/03/guided-deq-blog.html
|
<div class="padded-content">
<div class="title">Which valley, and how deep: training atomic relaxation at a fraction of the memory</div>
<!-- <div class="title">Which valley, and how deep: training atomic relaxation at a fraction of the memory</div> -->
<info datetime="">
Jul 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, on training machine-learned interatomic potentials to relax atomic structures — the expensive inner loop of materials discovery — without the usual accuracy-versus-memory trade-off.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/06/24/two-ways-to-slim-blog.html
|
<div class="padded-content">
<div class="title">Two ways to slim a model: remember vs recompute</div>
<!-- <div class="title">Two ways to slim a model: remember vs recompute</div> -->
<info datetime="">
Jun 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, on how to shrink a neural network without making it dumber — and why the right way to prune depends on what a given part of the model is actually doing.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/05/25/one-rank-at-a-time-blog.html
|
<div class="padded-content">
<div class="title">One rank at a time: cascading error dynamics in sequential learning</div>
<!-- <div class="title">One rank at a time: cascading error dynamics in sequential learning</div> -->
<info datetime="">
May 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our recent arXiv preprint — accepted to TMLR — with Mahtab Alizadeh Vandchali and Fangshuo (Jasper) Liao on error propagation in sequential rank-one learning.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/05/19/asymmetric-input-blog.html
|
<div class="padded-content">
<div class="title">One model, two roles: emergent specialization in a shared recurrent Transformer</div>
<!-- <div class="title">One model, two roles: emergent specialization in a shared recurrent Transformer</div> -->
<info datetime="">
May 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our recent arXiv preprint with Jucheng Shen and Barbara Su on whether a shared-weight recurrent Transformer can develop multiple internal roles on its own.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/05/14/adapad-blog.html
|
<div class="padded-content">
<div class="title">From PCA to LoRA: why fine-tuning could have been parallel all along</div>
<!-- <div class="title">From PCA to LoRA: why fine-tuning could have been parallel all along</div> -->
<info datetime="">
May 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our recent arXiv preprint with Barbara Su and Fangshuo (Jasper) Liao on parallel rank-one deflation for low-rank fine-tuning.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/05/08/ghost-mamba2-blog.html
|
<div class="padded-content">
<div class="title">GHOST: pruning Mamba2 by what each channel actually does</div>
<!-- <div class="title">GHOST: pruning Mamba2 by what each channel actually does</div> -->
<info datetime="">
May 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our ICML 2026 paper with Michael Menezes on state pruning for Mamba2 selective state-space models.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/05/02/multiplicative-gaussian-input-blog.html
|
<div class="padded-content">
<div class="title">How a little Gaussian dust changes how a network learns</div>
<!-- <div class="title">How a little Gaussian dust changes how a network learns</div> -->
<info datetime="">
May 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our recent paper with Afroditi Kolomvaki, Fangshuo (Jasper) Liao, Evan Dramko, and Ziyun Guang on training two-layer ReLU networks under multiplicative Gaussian input noise.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/04/24/stochastic-self-stabilization-blog.html
|
<div class="padded-content">
<div class="title">Why Stochastic Gradient Descent Stops Just Short of the Edge</div>
<!-- <div class="title">Why Stochastic Gradient Descent Stops Just Short of the Edge</div> -->
<info datetime="">
Apr 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our recent paper with Fangshuo (Jasper) Liao and Afroditi Kolomvaki on the Edge of Stochastic Stability.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/04/22/nesterov-acceleration-blog.html
|
<div class="padded-content">
<div class="title">Provable Acceleration of Nesterov's Momentum for Deep ReLU Networks</div>
<!-- <div class="title">Provable Acceleration of Nesterov's Momentum for Deep ReLU Networks</div> -->
<info datetime="">
Apr 2026
</info>
<span class="body">We published a new blog post on our AI-OWLS page, summarising our ALT 2024 paper with Fangshuo (Jasper) Liao on provable accelerated convergence of Nesterov’s momentum for deep ReLU neural networks.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/04/18/maxkcut-dsatur-blog.html
|
<div class="padded-content">
<div class="title">Spectral vs. Combinatorial for Max-3-Cut</div>
<!-- <div class="title">Spectral vs. Combinatorial for Max-3-Cut</div> -->
<info datetime="">
Apr 2026
</info>
<span class="body">We published a new blog post on our explore-quantum page, comparing spectral and combinatorial construction heuristics for Max-3-Cut.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/04/11/maxkcut-randr2-blog.html
|
<div class="padded-content">
<div class="title">Randomized Rank-2 for Max-3-Cut — When Two Eigenvectors Beat One</div>
<!-- <div class="title">Randomized Rank-2 for Max-3-Cut — When Two Eigenvectors Beat One</div> -->
<info datetime="">
Apr 2026
</info>
<span class="body">We published a new blog post on our explore-quantum page, introducing a 3-phase spectral pipeline for Max-3-Cut that outperforms simulated annealing on half of the graph families tested.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/04/05/maxkcut-rank1-blog.html
|
<div class="padded-content">
<div class="title">Rank-1 as a Building Block for Million-Node Max-3-Cut</div>
<!-- <div class="title">Rank-1 as a Building Block for Million-Node Max-3-Cut</div> -->
<info datetime="">
Apr 2026
</info>
<span class="body">We published a new blog post on our explore-quantum page, describing how three algorithmic insights make rank-1 spectral methods practical at extreme scale for Max-3-Cut.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/03/26/maxkcut-gpu-blog.html
|
<div class="padded-content">
<div class="title">What Can 15 Obsolete GPUs Do for Combinatorial Optimization?</div>
<!-- <div class="title">What Can 15 Obsolete GPUs Do for Combinatorial Optimization?</div> -->
<info datetime="">
Mar 2026
</info>
<span class="body">We published a follow-up blog post on our explore-quantum page, describing our GPU-parallel implementation of the low-rank Max-K-Cut solver.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2026/03/25/maxkcut-blog.html
|
<div class="padded-content">
<div class="title">Exploiting Low-Rank Structure in Max-K-Cut Problems</div>
<!-- <div class="title">Exploiting Low-Rank Structure in Max-K-Cut Problems</div> -->
<info datetime="">
Mar 2026
</info>
<span class="body">We published a new blog post on our explore-quantum page, describing our work on exploiting low-rank structure in Max-K-Cut problems.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2024/05/18/Update4.html
|
<div class="padded-content">
<div class="title">Major update</div>
<!-- <div class="title">Major update</div> -->
<info datetime="">
May 2024
</info>
<span class="body">Several updates happened in the past months: 1 paper was accepted at ICLR 2024 on adaptive step size for federated learning; 1 paper was accepted at ICML 2024 on the first work on how error propagates in deflation PCA; 1 paper was accepted at ISIT 2024 on a fast algorithm for FixMatch; and 1 paper was accepted at ICRA 2024.
This constitutes the update of OptimaLab for the past 5 months. More details are in the Publications webpage or Google Scholar.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2024/01/08/Update3.html
|
<div class="padded-content">
<div class="title">Major update</div>
<!-- <div class="title">Major update</div> -->
<info datetime="">
Jan 2024
</info>
<span class="body">Several updates happened in the past months: 1 paper was accepted at ICRC with the best poster award (Congrats, David Quiroga); 1 paper was accepted at ALT 2024; 1 paper was accepted at CPAL 2024; 1 paper accepted at NeurIPS 2023; 1 paper accepted at ICCV 2023; 1 paper accepted at ACML 2023; 1 journal accepted at the International Union of Crystallography Journal; and 1 journal paper accepted at the Machine Learning Journal (Springer).
This constitutes the update of OptimaLab for the past 5 months. More details are in the Publications webpage or Google Scholar.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2023/04/29/Update2.html
|
<div class="padded-content">
<div class="title">Major update</div>
<!-- <div class="title">Major update</div> -->
<info datetime="">
Apr 2023
</info>
<span class="body">OptimaLab has received an Amazon Research Award (ARA) and a Microsoft Research Award! Thank you sponsors!
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2023/04/29/Update.html
|
<div class="padded-content">
<div class="title">Major update</div>
<!-- <div class="title">Major update</div> -->
<info datetime="">
Apr 2023
</info>
<span class="body">3 papers accepted at AISTATS 2023; 1 paper accepted at ICASSP; 1 paper accepted at ICRA; 2 journal publications on GIST and fast quantum tomography:
This constitutes the update of OptimaLab the past 5 months. More details in the Publications webpage or Google Scholar.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2022/08/01/TMLR.html
|
<div class="padded-content">
<div class="title">Journal on neural network training accepted at Transactions on Machine Learning Research (TMLR)</div>
<!-- <div class="title">Journal on neural network training accepted at Transactions on Machine Learning Research (TMLR)</div> -->
<info datetime="">
Aug 2022
</info>
<span class="body">Fangshuo (Jasper) Liao has led the effort to prove why and when one can achieve this by iteratively creating, training, and combining randomly selected subnetworks in deep learning.
The journal is accepted at the Transactions on Machine Learning Research (TMLR) - details below.
</span>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/2022/06/26/UAI22.html
|
<div class="padded-content">
<div class="title">Two papers accepted at UAI 2022</div>
<div class="image" style="background-image:url('/public/amsterdam.jpeg')"></div>
<!-- post.photo_url -->
</div>
|
|
Анализировать url
|
|
/archive/
|
More...
|
|
Анализировать url
|