Papers of the day   All papers

What does it mean to understand a neural network?

Comments

hardmaru: What does it mean to understand a neural network? https://arxiv.org/abs/1907.06374 https://t.co/ta8OgYjNTj

2 replies, 473 likes


KordingLab: What does it mean to understand a neural network: https://arxiv.org/abs/1907.06374 - Tim Lillicrap and me try to think deeply about how pondering neural networks changes our goals as neuroscientists.

15 replies, 465 likes


KordingLab: Our paper "what does it mean to understand a neural network?" is deeply shaping the way I think about the future of neuroscience. https://arxiv.org/abs/1907.06374 Where should we publish it?

17 replies, 214 likes


hardmaru: I thought it was a pretty cool paper: https://twitter.com/hardmaru/status/1151018286416633857?s=21

4 replies, 70 likes


KordingLab: Physiologists: does our neural network based perspective ring true that development and plasticity as opposed to computation should be emphasized? https://arxiv.org/abs/1907.06374

15 replies, 49 likes


Lucas C Parra: A new consensus is forming ... The structure of neural networks can largely define their function making learning much easier; this structure may be encoded fairly compactly in genome/program code: https://www.nature.com/articles/s41467-019-11786-6 https://ai.googleblog.com/2019/08/exploring-weight-agnostic-neural.html https://arxiv.org/abs/1907.06374

1 replies, 18 likes


Guillermo Valle: Cool opinion paper. Understanding a trained net's knowledge is too hard, so we study the learning process. Why don't we do the same with brains? maybe that should be the current focus of neuroscience.

1 replies, 16 likes


KordingLab ๐Ÿ‘จโ€๐Ÿ’ป๐Ÿง โˆ‡๐Ÿ”ฌ๐Ÿ“ˆ,๐Ÿ‹๏ธโ€โ™‚๏ธโ›ท๏ธ๐Ÿ‚๐Ÿ›น๐Ÿ•บโ›ฐ๏ธโ˜•๐Ÿฆ–: @tyrell_turing @jeffrey_bowers @bradpwyble @SaraASolla @Federico_claudi @sam_neuro @TimKietzmann I just wanted to leave a link here to a paper where Tim Lillicrap and myself grapple with what understanding means to us: https://arxiv.org/abs/1907.06374

0 replies, 15 likes


Gary Lupyan: Indeed. The abstract is perhaps the best one-paragraph argument against the "core knowledge" strain of developmental psychology that I've seen.

2 replies, 11 likes


Venkat Ramaswamy: Lillicrap & @KordingLab in https://arxiv.org/abs/1907.06374 contemplate what it means to understand a neural net. They close asking how to prove the potential impossibility of understanding a neural net. With Turing Machines, we know how. (I also discuss it in <https://www.biorxiv.org/content/10.1101/639724v1>) https://t.co/M8PZuIiAca

1 replies, 9 likes


Gary Lupyan: Looking forward to discussing what makes Neural networks understandable https://arxiv.org/pdf/1907.06374.pdf at an upcoming meeting (https://lucid.wisc.edu/hamlet/). But @KordingLab, that tic tac toe strategy isn't right, is it? :E https://t.co/BaD4u5yBG0

1 replies, 6 likes


๐™ถ๐šž๐š’๐š•๐š•๐šŠ๐šž๐š–๐šŽ ๐™ณ๐šž๐š–๐šŠ๐šœ ๐Ÿค–๐Ÿ”๐Ÿง : What does it mean to understand a #NeuralNetwork?https://arxiv.org/abs/1907.06374 https://t.co/LjUG0DC6qm

0 replies, 4 likes


IntuitionMachine: "Instead of asking how the brain works we should, arguably, ask how it learns to work." https://arxiv.org/abs/1907.06374 @KordingLab

1 replies, 4 likes


Jon Bloom: "However, while we may not be very hopeful about communicating a working model of the dynamics, we can still hope to communicate how biology, nature, sets up the brain...While learned information is unlikely to be compressible, we may at least hope that biological information is"

0 replies, 4 likes


Kevin Mitchell: "The analogy suggests that neuroscience would benefit from a focus on learning and development." ๐Ÿ˜€๐Ÿ‘

0 replies, 4 likes


ๆณจ็›ฎใฎๆœ€ๆ–ฐarXivใ€ๆฏŽๆ—ฅๆ›ดๆ–ฐใ€‘: 2019/07/15 ๆŠ•็จฟ 3ไฝ LG(Machine Learning) What does it mean to understand a neural network? https://arxiv.org/abs/1907.06374 7 Tweets 15 Retweets 71 Favorites

0 replies, 3 likes


Fred Crevecoeur: @KordingLab and Tim Lilicrap eloquently advancing our ideas about what it means to understand the brain https://arxiv.org/abs/1907.06374

1 replies, 3 likes


Arip Asadulaev: This Tractatus is an answer to Timothy P. Lillicrap and Konrad P. Kording @KordingLab paper https://arxiv.org/pdf/1907.06374.pdf. https://philpapers.org/archive/IERWDI.pdf https://t.co/WappmdlDFK

1 replies, 3 likes


SB: What does it mean to understand a neural network? #DeepLearning #MachineLearning via http://twinybots.ch https://arxiv.org/abs/1907.06374

0 replies, 3 likes


Tristan Bekinschtein: The supremacy of learning and development to understand neural function. @DrVLeong @DuncanAstle

1 replies, 2 likes


๐–†๐–Ž๐–œ๐–†๐–—๐–‘๐–”๐–—๐–‰: What does it mean to understand Neural Networks by @KordingLab? Such networks are thus dramatically easier to understand in terms of the code that makes them than the resulting properties, such as tuning or connections. arXiv: https://arxiv.org/abs/1907.06374

0 replies, 2 likes


KordingLab ๐›โˆˆ๐Ÿง : @stochastician I guess the question is what we mean with understand: https://arxiv.org/abs/1907.06374 - but I agree, in many ways we do not understand ANNs and never will.

2 replies, 2 likes


Cian McCafferty: Wow, this piece has some fascinating insights and suggestions: that components & principles of neural development may be much more decodable, *and therefore useful to study*, than the incredibly complex resulting connections & weights.

0 replies, 1 likes


Hongbiao: Yeah! Discuss the question: when do we can say we understand a neural network?

0 replies, 1 likes


BaThiaMkWr: What does it mean to understand a neural network? Timothy P. Lillicrap and Konrad P. Kording: https://arxiv.org/abs/1907.06374 #artificialintelligence #deeplearning #machinelearning https://t.co/Pule4QYJTt

0 replies, 1 likes


Cornichecorp Digital Solutions: What does it mean to understand a neural network? #DeepLearning #MachineLearning via http://twinybots.ch https://arxiv.org/abs/1907.06374

0 replies, 1 likes


Marco Manca: MT @KordingLab Can a focus on development help us further our understanding of neuroscience? Musings inspired by experiences on machine learning... https://arxiv.org/abs/1907.06374

0 replies, 1 likes


KordingLab: @AdrianAWanner hard to know. If I were to take the opposite argument I would argue that the connectome may mostly be made by learning. And can thus have far more entropy than the DNA. https://arxiv.org/abs/1907.06374

1 replies, 0 likes


Content

Found on Jul 16 2019 at https://arxiv.org/pdf/1907.06374.pdf

PDF content of a computer science paper: What does it mean to understand a neural network?