How can a system change
without losing what matters?
I’m Navish Kumar, a machine-learning researcher at the University of Basel.
I study how learning systems change—and build tools that make that change inspectable.
Follow the workA body of work, in motion.
Scroll through the work. Enter an idea to follow it further.
PastFoundations
From relationships in networks to the geometry of learning.
- 2020 · Published research
Interaction dynamics
Look at who replies to whom—not just how much each group posts.
Enter the idea - 2021 · Published · Linear Algebra and its Applications
Extremal spectral bounds
Turn a spectral warning into a lower bound on structural repair.
Enter the idea - 2022 · Published · American Journal of Combinatorics
Normalized gain Laplacians
Change one relationship. Watch the inconsistency enter the spectrum.
Enter the idea - 2022–2023 · NeurIPS 2022 workshop · 2023 preprint
Urban micro-regions
Change the street context. See parking, travel and walking add up.
Enter the idea - 2025 · Preprint
Natural-gradient guarantees
Fit a distribution, not just a point—and expose the update geometry.
Enter the idea
NowActive work
What should change, what should stay, and who gets to decide?
- 2026 · Active research
Experience Replay
Learn something new. Lose an old skill. Choose the missing correction.
Enter the idea - 2026 · Active research
Rank Feasibility
Increase the available rank. Separate possible from affordable.
Enter the idea - 2026 · Replication / ongoing investigation
TiC-LM replication
Advance time. Watch the same historical data become stale.
Enter the idea - 2026 · System in development
CasePath
Find the contradiction. Refuse the action. Correct only what depends on it.
Enter the idea
FrontierDirection
Intelligence that can act in a world we can inspect and revise.
Let’s talk
An idea worth understanding.
A system worth building.
navish.kumar@unibas.ch Research, collaboration, and difficult questions welcome.