Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Predictive coding has been unified with backpropagation

lesswrong.com

Post reply on HN
Print view View original

71–80 of 85 posts

Re: Predictive coding has been unified with backpropagation

#71

Post by TaupeRanger »

andyxor wrote: post #57
TaupeRanger wrote: post #53

Earlier quoted context omitted.

No. We really have no idea what is going on. We only know some basic psychology about it (holding 7 things in short term, etc.) If we knew something about implementation, we could implement human-like memory.

I suggest starting with the works by Howard Eichenbaum on memory and Edvard & May-Britt Moser (and John O'Keefe and Lynn Nadel) on place & grid cells. For the latest and greatest see https://twitter.com/doellerlab https://twitter.com/KordingLab https://twitter.com/preston_lab https://twitter.com/memorylab https://twitter.com/ptoncompmemlab https://twitter.com/MillerLabMIT https://twitter.com/hugospiers Once you start…

Literally nothing you posted surprises me in the least, and literally none of this work shows that we know anything at all about how memory is implemented. Perhaps read some of the many takedowns of so called "grid cells" which show that it is completely unsurprising and not at all interesting or noteworthy that activity in some parts of the brain correlates with location information. The important questions always remain unanswered.

Re: Predictive coding has been unified with backpropagation

#72

Post by klmadfejno »

TheRealPomax wrote: post #33
klmadfejno wrote: post #31

Earlier quoted context omitted.

I'm trying to imagine how that works. Imagine you've got a nueral net. One node identifies the number of feet. One node identifies that number of wings. One node identifies color. This feeds into a layer that tries to predict what animal it is. With backprop, you can sort of assume that given enough scale your algo will identify these important features. With local learning, wouldn't you get a tendency to identify th…

The fallacy there is the idea that "one node" does anything useful, rather than optimizing itself in a way that you have _no idea_ what it actually codes for, but at the emergent level, you see it contribute to coding for wing detection, or color detection, or more likely actually seventeen different things that are supposedly unrelated, it just happens to be generating values that somehow contribute to a result for…

That's, at best, pedantically true. You can determine the function of individual components of a network, and they will correspond to concrete thing. It's just that the utility of doing this is low in the scheme of things and the function of individual components is going to be fuzzier than nice constructs that humans like to think in. If you wanted to take painstaking steps to align the functionality of nodes to identify specific features per the example, you could do this and the network would work just fine. It's an appropriate way to simplify an explanation of how the model works.

Re: Predictive coding has been unified with backpropagation

#73

Post by klmadfejno »

SamBam wrote: post #41
klmadfejno wrote: post #31

Earlier quoted context omitted.

I'm trying to imagine how that works. Imagine you've got a nueral net. One node identifies the number of feet. One node identifies that number of wings. One node identifies color. This feeds into a layer that tries to predict what animal it is. With backprop, you can sort of assume that given enough scale your algo will identify these important features. With local learning, wouldn't you get a tendency to identify th…

> Is there a need for a sort of middleman like a one arm bandit kind of thing that makes a decision to spawn and despawn child nodes to explore the space more? What's the one-armed bandit? (Besides a slot machine.) My knowledge of this field is rusty, but I actually wrote my MSc thesis on novel ways to get Genetic Algorithms to more efficiently explore the space without getting stuck, so it sounds up my alley.

Haha sorry I meant multi arm bandit, which I'd presume you're familiar with.

Although I guess a single arm bandit would be something akin the secretary problem.

Re: Predictive coding has been unified with backpropagation

#74

Post by blueyes »

ShamelessC wrote: post #10
andyxor wrote: post #7

Earlier quoted context omitted.

there is no evidence of back-propagation in the brain. See Professor Edmund T. Rolls books on biologically plausible neural networks: "Brain Computations: What and How" (2020) https://www.amazon.com/gp/product/0198871104 "Cerebral Cortex: Principles of Operation" (2018) https://www.oxcns.org/b12text.html "Neural Networks and Brain Function" (1997) https://www.oxcns.org/b3_text.html

"There is just one problem: [biological neural networks] are physically incapable of running the backpropagation algorithm." From the linked article.

I read that sentence. The article is not the only source of truth in brain function, and its author may be too certain about the brain. In any case, there will always be dissimilarities between biological neurons and computations on silicon, which probably shouldn't be called neurons, in order to avoid confusion.

Re: Predictive coding has been unified with backpropagation

#75

Post by cs702 »

YeGoblynQueenne wrote: post #38
cs702 wrote: post #13

EDIT: Before you read my comment below, please see https://news.ycombinator.com/item?id=26702815 and https://openreview.net/forum?id=PdauS7wZBfC for a different view. -- If the results hold, they seem significant enough to me that I'd go as far as saying the authors of the paper would end up getting an important award at some point, not just for unifying the fields of biological and artificial intelligence , but also…

Note that the paper was rejected for publication in ICLR 2021: https://openreview.net/forum?id=PdauS7wZBfC

Yes, I linked to that same page at the top of my comment :-)

Re: Predictive coding has been unified with backpropagation

#76

Post by TheRealPomax »

klmadfejno wrote: post #72
TheRealPomax wrote: post #33

Earlier quoted context omitted.

The fallacy there is the idea that "one node" does anything useful, rather than optimizing itself in a way that you have _no idea_ what it actually codes for, but at the emergent level, you see it contribute to coding for wing detection, or color detection, or more likely actually seventeen different things that are supposedly unrelated, it just happens to be generating values that somehow contribute to a result for…

That's, at best, pedantically true. You can determine the function of individual components of a network, and they will correspond to concrete thing. It's just that the utility of doing this is low in the scheme of things and the function of individual components is going to be fuzzier than nice constructs that humans like to think in. If you wanted to take painstaking steps to align the functionality of nodes to ide…

That's literally what we cannot do in a neural net.

It's why they're so problematic: you can determine the propagation functions of individual nodes perfectly, and that knowledge tells you exactly nothing about all of the many things it's values contribute to. There is no "concrete thing" at the node level: a single node fundamentally can't see a wing, or a color, or anything else, that's only emergent behaviour of node constellations, and one node can contribute to many constellations simultaneously.

Heck, there often isn't even a "concrete thing" at the many of the constellation levels, concrete things don't start to emerge until you're looking at the full state of all end nodes.

Re: Predictive coding has been unified with backpropagation

#77

Post by cs702 »

nl wrote: post #61
cs702 wrote: post #13

EDIT: Before you read my comment below, please see https://news.ycombinator.com/item?id=26702815 and https://openreview.net/forum?id=PdauS7wZBfC for a different view. -- If the results hold, they seem significant enough to me that I'd go as far as saying the authors of the paper would end up getting an important award at some point, not just for unifying the fields of biological and artificial intelligence , but also…

I don’t think anyone familiar with the field is in anyway surprised by this results. The breakthrough seems really limited to showing it holds for graphs. We already knew this was practically true though anyway.

Agree, no one is surprised.

But the authors successfully show how to train CNNs, RNNs, and LSTM RNNs without backpropagation, i.e., every layer learning only via local rules, without having to wait for gradients to be backpropagated to all layers before the entire model can move on to the next sample.

As I understand it, this work has paved a path for training very large networks in massively parallel, fully distributed hardware -- in the not too distant future.

Re: Predictive coding has been unified with backpropagation

#78

Post by YeGoblynQueenne »

cs702 wrote: post #75
YeGoblynQueenne wrote: post #38

Earlier quoted context omitted.

Note that the paper was rejected for publication in ICLR 2021: https://openreview.net/forum?id=PdauS7wZBfC

Yes, I linked to that same page at the top of my comment :-)

Oops, sorry- I noticed the link but I thought it was a HN url, like the one before... and I wondered why it was greyed-out (visited). But still I didn't check it out. My very bad.

Re: Predictive coding has been unified with backpropagation

#79

Post by selimthegrim »

adamnemecek wrote: post #36
selimthegrim wrote: post #35

Earlier quoted context omitted.

What were you going to say about Young tableaux?

Dynamic programming and reinforcement learning are just diagonalizations of the Young tableau. This is related to the spectral theorem.

Interesting. Is there some concrete realization you had in mind?

Re: Predictive coding has been unified with backpropagation

#80

Post by eli_gottlieb »

andyxor wrote: post #57
TaupeRanger wrote: post #53

Earlier quoted context omitted.

No. We really have no idea what is going on. We only know some basic psychology about it (holding 7 things in short term, etc.) If we knew something about implementation, we could implement human-like memory.

I suggest starting with the works by Howard Eichenbaum on memory and Edvard & May-Britt Moser (and John O'Keefe and Lynn Nadel) on place & grid cells. For the latest and greatest see https://twitter.com/doellerlab https://twitter.com/KordingLab https://twitter.com/preston_lab https://twitter.com/memorylab https://twitter.com/ptoncompmemlab https://twitter.com/MillerLabMIT https://twitter.com/hugospiers Once you start…

We know a fair bit about how cognitive maps work in 2D and 3D Euclidean environments. We know damn little about how nontrivial manifold structure can be learned, particularly in spaces with more than three dimensions.
Post reply on HN