frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Ask HN: Where does modern geometry survive contact with SGD?

1•ternaus•1h ago
Over the past year I worked through Frankel’s “The Geometry of Physics” cover to cover, not to relearn physics, but to rebuild a modern geometric toolbox as it is actually used there: manifolds, differential forms, connections and curvature, Lie groups and algebras, fiber bundles, gauge structure, and variational principles.

The motivation was practical rather than theoretical:

Which of these geometric structures, if any, actually survive discretization, noise, and SGD-style training in modern machine learning?

In physics, global and coordinate-free formulations were not aesthetic choices; they were forced when local reasoning stopped working. A recurring structural pattern was:

structure -> symmetry -> invariance -> dynamics -> observables

In modern ML we increasingly see analogous issues:

* parameter symmetries and large quotient spaces * non-Euclidean data (graphs, meshes, manifolds) * highly structured hypothesis classes * training dynamics that are not well-described by flat Euclidean optimization

Some geometric ideas clearly paid off (e.g. equivariance via group actions). Many others did not. I’m trying to understand where future leverage might still lie, and where geometry collapses to interpretation or preconditioning.

Below is my current (incomplete) map of where modern geometry already shows up in ML, or plausibly could.

1. Geometry of data (base spaces)

Manifolds, stratified spaces, graphs and meshes; discrete differential operators (graph Laplacians, discrete Hodge theory); topological summaries (persistent homology).

This seems strongest for representation, spectral methods, and diagnostics. The open question is how much of this geometry can couple dynamically to training, rather than remain preprocessing or analysis.

2. Geometry of hypothesis spaces (architectures)

So far the most successful direction:

* symmetry and equivariance via group actions * quotienting hypothesis spaces * convolution as representation theory * SE(3)- / gauge-equivariant models * architectures encoding invariants or conservation laws

Here geometry restricts the hypothesis class before optimization. I suspect there is still room beyond global groups, toward local gauge structure, fiber bundle–valued representations, and architectures defined by connections rather than coordinates.

3. Geometry of parameters and optimization

Optimization on manifolds (Stiefel, Grassmann, SPD cones), structured or low-rank parameterizations, information geometry and natural gradients.

This seems most effective when constraints are hard and geometric. In looser settings, much of this reduces to preconditioning. It’s unclear where deeper geometric structure still matters at scale.

4. Geometry of training dynamics

Viewing training as a stochastic dynamical system:

* gradient descent as discretized flow * SGD as an SDE * trajectories on manifolds * attractors and metastability

This connects to dynamical systems, stochastic analysis, and geometric mechanics, but remains underdeveloped relative to its apparent relevance.

5. Discrete vs smooth geometry

Modern ML is deeply discrete: finite precision, quantization, sparse activations, graph-based computation. Smooth differential geometry may be the wrong limit in some regimes. Discrete differential geometry or combinatorial curvature might be more appropriate.

Some failures of “geometric ML” may simply be failures of choosing the wrong geometric category.

What I’m looking for:

* geometric structures that have actually influenced model or optimizer design beyond equivariance * where Riemannian or information-geometric ideas help in large-scale training * which geometric frameworks seem promising but currently mismatched with SGD * directions I’m missing

Perspectives from theory groups, applied math, and industry research labs would be very welcome.

TWIRL Integer Factorization

https://en.wikipedia.org/wiki/TWIRL
1•walterbell•55s ago•0 comments

Agentic search (glob/grep/read) works better than RAG and vector DB

https://twitter.com/dani_avila7/status/2018766464933613871
1•stared•2m ago•0 comments

The €10 Mirror: Why Enterprise Security Looks Like a Kid's Toy

https://labs.itresit.es/2026/02/04/the-e10-mirror-why-enterprise-security-looks-like-a-kids-toy/
1•Yippee-Ki-Yay•3m ago•0 comments

How does OpenAI balance long-term research bets with product-forward research?

https://twitter.com/markchen90/status/2018779039205667046
1•tosh•4m ago•0 comments

Public Notice: I Am Your AIB and the Warning That Came True

1•rowanseerwald•6m ago•0 comments

voyage-multimodal-3.5: a new multimodal retrieval frontier with video support

https://blog.voyageai.com/2026/01/15/voyage-multimodal-3-5/
1•fzliu•8m ago•0 comments

Detecting and Monitoring OpenClaw (clawdbot, moltbot) in your environment

https://isc.sans.edu/diary/32678
1•Binary_Impact•11m ago•1 comments

Quantum Computing for Programmers

https://github.com/qcc4cp/qcc
1•altro•12m ago•0 comments

FireClaw: Personal OpenClaw assistant in a single binary, built on Firecracker

https://github.com/AFK-surf/fireclaw
2•Johnson8053•14m ago•1 comments

Melinda French Gates Appears to Confirm Divorce Was Related to Epstein

https://gizmodo.com/melinda-french-gates-appears-to-confirm-divorce-with-bill-was-related-to-epst...
3•petethomas•15m ago•0 comments

Show HN: Sentinel – a Pingora-based reverse proxy (inspired by River)

https://sentinel.raskell.io/
3•raskelll•18m ago•0 comments

Ask HN: Does anyone keep prompts and reasoning as part of dev cycle?

1•sshadmand•22m ago•1 comments

Russian spy spacecraft have intercepted Europe's key satellites

https://www.ft.com/content/cd08c49c-658e-49c9-9a15-234f2bfc2074
3•mraniki•23m ago•2 comments

The world is more equal than you think

https://economist.com/graphic-detail/2026/02/03/the-world-is-more-equal-than-you-think
6•andsoitis•25m ago•2 comments

A new nuclear arms race beckons

https://economist.com/international/2026/02/03/a-new-nuclear-arms-race-beckons
1•andsoitis•27m ago•0 comments

Rust Is Just a Tool

https://lewiscampbell.tech/blog/260204.html
2•LAC-Tech•27m ago•0 comments

Show HN: Augmenting developer docs into high-level interactive mental models

https://docmaps-web.vercel.app/
3•b_mutea•28m ago•0 comments

Elon Musk's mega-merger makes little sense business sense

https://www.economist.com/business/2026/02/03/elon-musks-mega-merger-makes-little-business-sense
2•andsoitis•29m ago•0 comments

How to squeeze a lexicon (2001) [pdf]

https://marcinciura.wordpress.com/wp-content/uploads/2019/10/lexicon.pdf
1•mci•30m ago•0 comments

Ask HN: At what point in the future will AIs stop working or not be correct?

2•roschdal•36m ago•1 comments

Show HN: TitanShell – Security-first desktop client for OpenClaw

https://github.com/DaguangZhou/TitanShell
1•snowwolf_zdg•37m ago•1 comments

Show HN: Gateway – An open-source proxy to securely handle BYOK keys

https://github.com/glueco/gateway
3•mumernisar•42m ago•1 comments

Echoes

https://thelehrhaus.com/culture/echoes/
2•barry-cotter•43m ago•0 comments

Manna by Marshall Brain

https://marshallbrain.com/manna1
2•jpmitchell•45m ago•0 comments

CUBO the Industrial-Grade Local RAG

https://github.com/PaoloAstrino/cubo
3•50kIters•45m ago•0 comments

Show HN: Astrolabe – Navigate Your Data Universe in Nextcloud

https://blog.coutinho.io/introducing-astrolabe-navigate-your-data-universe-in-nextcloud
2•cbcoutinho•52m ago•0 comments

"Superhuman": A 13 year old boy swims 2.5 miles to save family swept out to sea

https://www.cbsnews.com/news/boy-swims-hours-save-mom-siblings-swept-out-sea-superhuman/?ftag=YHF...
7•gurjeet•52m ago•0 comments

Oracle to raise $50B as AI debt piles up

https://www.marketwatch.com/story/oracles-monster-25-billion-debt-financing-points-to-anxieties-a...
4•zerosizedweasle•54m ago•0 comments

Show HN: Yutovo – visual online and desktop calculator inside a text editor

https://yutovo.com
3•denprog•55m ago•0 comments

Siemens Energy Bets $1B That A.I. Power Demand Will Last

https://www.nytimes.com/2026/02/03/business/energy-environment/siemens-energy-ai-power-demand.html
2•ChrisArchitect•1h ago•1 comments