HackerTube

Geoffrey Hinton talk "What is wrong with convolutional neural nets ?"

trwappers2 HN recommenders · 2017–2018
200K views · lecture · dated · Uploaded 2017-04-03

Geoffrey Hinton gives an MIT talk on what is wrong with convolutional neural networks and introduces his capsules project. The description points to the capsule network papers.

Recommended as Hinton's own explanation of capsule networks and his critique of pooling in CNNs.

Machine learning & deep learningComputer vision, speech & NLP

Watch on YouTube

5 comments from Hacker News

@bra-ket2017 · in “Is AI Riding a One-Trick Pony?”

Prof. Hinton has an interesting talk about his new 'capsule' model based on psychology of shape perception: [VIDEO]

RecommendsView on HN
@uoaei2017 · in “BadNets: Identifying vulnerabilities in the machine learning model supply chain”

I have a hypothesis that these errors in interpretation stem mainly from the max pooling layers that occurs in nearly all CNNs today. Simply cutting out a large fraction (at least 75%!) of the information passing through each pooling layer is making the models quite brittle. That, and obviously it would be a big help to be able to interpret meaningful symbols (e.g. words) from the images rather than just recognize a certain arbitrary class of pattern activations. That is pretty hard, of course, but IMO essential for the upcoming self-driving car revolution, if it is to succeed. G. Hinton has some ideas though they aren't very fleshed out yet.[1] [1] [VIDEO]

CitesView on HN
@dr_zoidberg2017 · in “A New Twist on Neural Networks”

Hinton explains the concept of capsules in this video: [VIDEO] Which is a lot better than reading someone tell you about this new idea that is called "capsules" but doesn't go into detail. The only thing is that, when this presentation was given, it seems they hadn't worked much more than MNIST (so the new thing now would be the toys-recognition net). Better source, with date: [LINK] (December 2014, for the lazy).

RecommendsView on HN
@giardini2017 · in “Ask HN: What's your favorite Machine Learning topic?”

Hinton's "capsule networks": [LINK] [VIDEO] [LINK] They sound a bit like "grandmother neurons/cells": [LINK] I would very much appreciate an ELI5 of capsules and some prognostications about how capsules may affect the state of the art in neural networks and the current flood of papers and graduate students in the pipeline of ML.

CitesView on HN
@nabla92018 · in “Deciphering China’s AI Dream”

I'm certain that AI business will plateau and there will be yet another AI winter without new big breakthroughs in fundamental research. We are currently living in a era where the results from Hinton & Canadian Mafia are still begin adopted and refined. It's just alternating layers of affine transformations and nonlinearity with lots of tricks and improved routing. There needs to be fundamental theoretical breakthrough in every decade or so, and not just refinements and new applications. Hinton made his "What is wrong with convolutional neural nets" speech years ago " [VIDEO] I believe that Hinton's capsule networks type attempts are what we need more. They may not deliver immediately impressive results because fundamental research involves lots of setbacks.

CitesView on HN