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. Lindbladians: Quoting the Wikipedia article: "[...] is one of the general forms of Markovian master equations describing open quantum systems. It generalizes the Schrödinger equation to open quantum systems; that is, systems in contact with their surroundings." It's not a term I was familiar with, as opposed to the Schrödinger equation, which I believe might be related to quantum mechanics moving into increasingly applied contexts: the shift between theoretical physics and engineering is seldom possible without sacrificing a lot of the simplifying assumptions that made the physics manageable to begin with.
Some recent articles:
- Learning and certification of local time-dependent quantum dynamics and noise
- Fast-forwardable Lindbladians imply quantum phase estimation
- Multi-block exceptional points in open quantum systems
- Simulation of bilayer Hamiltonians based on monitored quantum trajectories
- Exponential Lindbladian fast forwarding and exponential amplification of certain Gibbs state properties
2. ZKPs: Zero-knowledge proofs, i.e. How To Prove You Know Something Without Saying It. As you can imagine, it's a powerful concept in cryptography — so far more frequently used as a theoretical building block than widely deployed — but if you haven't encountered it before it's quite mind-blowing.
Some recent articles:
- Privacy-Preserving On-chain Permissioning for KYC-Compliant Decentralized Applications
- VDDP: Verifiable Distributed Differential Privacy under the Client-Server-Verifier Setup
- Optimizing Privacy-Preserving Primitives to Support LLM-Scale Applications
- ZKProphet: Understanding Performance of Zero-Knowledge Proofs on GPUs
- Confidentiality-Preserving Verifiable Business Processes through Zero-Knowledge Proofs
3. Algonauts 2025 Challenge: From the website: The Algonauts Project 2025 challenge will evaluate computational models on how well they predict human brain data recorded while humans perceive multimodal naturalistic movies, using CNeuroMod, a massive human brain dataset collected for that purpose. The challenge is already concluded, and you can read about the outcomes in the papers below. I'm generally skeptical of the more grandiose claims about neurotechnology: just skimming an introductory graduate-level textbook on the brain and one or two reviews of the SOTA in electrodes and other sensors will give you an idea how much complexity is going own in there (and how much of it we know we don't know) and how limited is our window into it. On the other hand, I'm simultaneously awe-struck by how much we have learned and how much our sensors have advanced! As long as you keep your expectations closer to science than to business-fiction there's a lot to learn and be impressed by.
Some recent articles:
- Stacked Regression using Off-the-shelf, Stimulus-tuned and Fine-tuned Neural Networks for Predicting fMRI Brain Responses to Movies (Algonauts 2025 Report)
- Insights from the Algonauts 2025 Winners
- Predicting Brain Responses To Natural Movies With Multimodal LLMs
- Multimodal Recurrent Ensembles for Predicting Brain Responses to Naturalistic Movies (Algonauts 2025)
- VIBE: Video-Input Brain Encoder for fMRI Response Modeling