A few of the ideas, topics, and commonplaces that have been gaining steam on arXiv during the last few months (yes, I do a lot of filtering to keep away most of the genAI stuff; explainer)
1. principled foundation: The term is self-explaning and points towards the sort of robust, generailzable techniques that make a lot of the flashier work possible (e.g. I should learn — and talk about — more about high-end chip design software).
Some recent articles:
- BeePL: Correct-by-compilation kernel extensions
- A statistical physics framework for optimal learning
- Why Neural Network Can Discover Symbolic Structures with Gradient-based Training: An Algebraic and Geometric Foundation for Neurosymbolic Reasoning
2. quasiparticle picture: I don't know remotely enough about quantum physics to read these papers, but at the very least the Wikipedia article is fascinating, and the idea is suggestive in general.
Some recent articles:
- Time Evolution of the Symmetry Resolved Entanglement Entropy after a Mass Quench
- $ν$-QSSEP: A toy model for entanglement spreading in stochastic diffusive quantum systems
- Integrability and charge transport in asymmetric quantum-circuit geometries
- A simpler probe of the quantum Mpemba effect in closed systems
- Subsystem Information Capacity in Random Circuits and Hamiltonian Dynamics
3. content diversity: Far from a new problem (would/will hypothetical chatbot-first interfaces to the Internet make it better, worse, structurally different?) so it's always nice to see people working on it. The first paper comes from Google people: it's a solid, interesting idea, but one that doesn't seem to fit Google's current strategy and practices. It would be straightforward to implement on arbitrary large and fast-growing item collections, so perhaps somebody will do it in some prosocial context.
Some recent articles:
- Item-centric Exploration for Cold Start Problem
- Simulating User Watch-Time to Investigate Bias in YouTube Shorts Recommendations
- Patterns and Dynamics of Netflix TV Show Popularity
- Bayesian-Guided Diversity in Sequential Sampling for Recommender Systems