Perhaps I'm seeking an "AlphaGo moment" for Active inference, where using the framework leads to building algorithms that are on par with or surpass humans on some task(s).
The Active Inference man himself! Thank you so much for replying! After going through the article, it still didn't address what SOTA benchmarks we're solving with ActInf. Having an explainable physics discovery is very insightful, but how do you aim to benchmark it?
Does Active Inference offer only an elegant theory and interpretation of existing Bayesian techniques or an alternative to Reinforcement Learning? Has it demonstrated empirical success over existing techniques on RL tasks? What's all the fuss about?
How does Popper come in here?
Ahh this looks like an interesting read, thanks for sharing!
There's a CompNeuro one: go.bsky.app/7VFUkdn
Thank you so much for sharing this!
It's been about 1.5 months since I started my Masters in Computational Neuroscience right after graduating from a CS undergrad. I'm interested in working in NeuroAI stuff, what books and papers are the best to get started with it? Would love to hear from CompNeuro people here.
