Predictive coding has been unified with backpropagation • Hacker News Board

Re: Predictive coding has been unified with backpropagation

#31

Post by klmadfejno » Mon, Apr 05, 2021, 6:41 PM UTC

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'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 the easily identifiable features many times? 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?

Re: Predictive coding has been unified with backpropagation

#32

Post by tsmithe » Mon, Apr 05, 2021, 6:42 PM UTC

adamnemecek wrote: post #24

I think that this sort of forward backward thing is a very general idea. There’s a one to many relationship called the adjoint, and a many to one relationship called the norm. I wrote something about this here https://github.com/adamnemecek/adjoint

In fact, the compositional structure underlying that of predictive coding [0,1] is abstractly the same as that underlying backprop [2]. (Disclaimer: [0,1] are my own papers; I'm working on a more precise and extensive version of [1] right now!)

[0] https://arxiv.org/abs/2006.01631 [1] https://arxiv.org/abs/2101.10483 [2] https://arxiv.org/abs/1711.10455

Re: Predictive coding has been unified with backpropagation

#33

Post by TheRealPomax » Mon, Apr 05, 2021, 6:46 PM UTC

klmadfejno wrote: post #31

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'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 the features the various constellations detect.

(meaning it might also actually cause one or more constellations to perform worse than if it wasn't contributing, and realistically, you'll never know)

Re: Predictive coding has been unified with backpropagation

#34

Post by phreeza » Mon, Apr 05, 2021, 6:56 PM UTC

This was already shown for MLPs some years ago, and it is not really that surprising that it applies to many other architectures. Note that while learning can take place locally, it does still require an upward and downward stream of information flow, which is not supported by the neuroanatomy in all cases. So while it is an interesting avenue of research, I don't think it's anywhere near as revolutionary as this blog post makes it out to be.

Re: Predictive coding has been unified with backpropagation

#35

Post by selimthegrim » Mon, Apr 05, 2021, 7:28 PM UTC

adamnemecek wrote: post #24

I think that this sort of forward backward thing is a very general idea. There’s a one to many relationship called the adjoint, and a many to one relationship called the norm. I wrote something about this here https://github.com/adamnemecek/adjoint

What were you going to say about Young tableaux?

Re: Predictive coding has been unified with backpropagation

#36

Post by adamnemecek » Mon, Apr 05, 2021, 7:36 PM UTC

selimthegrim wrote: post #35

adamnemecek wrote: post #24

I think that this sort of forward backward thing is a very general idea. There’s a one to many relationship called the adjoint, and a many to one relationship called the norm. I wrote something about this here https://github.com/adamnemecek/adjoint

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.

Re: Predictive coding has been unified with backpropagation

#37

Post by salawat » Mon, Apr 05, 2021, 7:37 PM UTC

nerdponx wrote: post #14

hctaw wrote: post #11

I don't know enough about biology or ML to know if what I'm posting below is totally wrong, but here goes. "Backprop" == "Feedback" of a non-linear dynamical system. Feedback is mathematical description of the behavior of systems, not a literal one. I don't know of BNNs are incapable of backprop anymore than an RLC filter is incapable of "feedback" when analyzing the ODE of the latter tells you that there's a feedbac…

The article says this: > The backpropagation algorithm requires information to flow forward and backward along the network. But biological neurons are one-directional. An action potential goes from the cell body down the axon to the axon terminals to another cell's dendrites. An axon potential never travels backward from a cell's terminals to its body. The point of the research here is that backpropagation turns out…

So... I don't understand.

>An action potential goes from the cell body down the axon to the axon terminals to another cell's dendrites.

How do you figure that doesn't allow backprop?

A neuronal bit is a loop of neurons. Information absolutely can back- propagate. If it couldn't, how does anyone think it'd be at all possible to learn how to get better at anything?

Neuron fires dendrite to axon, secondary neuron fires dendrite to Axon, Axon branches back to previous neuron's dendrites, rinse, repeat, or add more intervening neurons... Trying to disinclude backprop based on the morphology of a single neuron is... Kinda missing the point.

It's all about the level of connection between neurons and how long or whether a signal returns unmodified to the progenitor that effects the stability of the encoded information or behavior. At least to the best I've been able to plausibly model it. Haven't exactly figured out how to shove a bunch of measuring sticks in there to confirm or deny, but I just can't how a uniderectional action potential forwarding element implies lack of backprop in a graph of connections fully capable of developing cycles.

Re: Predictive coding has been unified with backpropagation

#38

Post by YeGoblynQueenne » Mon, Apr 05, 2021, 7:38 PM UTC

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

Re: Predictive coding has been unified with backpropagation

#39

Post by candiodari » Mon, Apr 05, 2021, 7:51 PM UTC

nerdponx wrote: post #14

hctaw wrote: post #11

I don't know enough about biology or ML to know if what I'm posting below is totally wrong, but here goes. "Backprop" == "Feedback" of a non-linear dynamical system. Feedback is mathematical description of the behavior of systems, not a literal one. I don't know of BNNs are incapable of backprop anymore than an RLC filter is incapable of "feedback" when analyzing the ODE of the latter tells you that there's a feedbac…

The article says this: > The backpropagation algorithm requires information to flow forward and backward along the network. But biological neurons are one-directional. An action potential goes from the cell body down the axon to the axon terminals to another cell's dendrites. An axon potential never travels backward from a cell's terminals to its body. The point of the research here is that backpropagation turns out…

Yeah, but then you run into the problem of computation speed. Any given neuron in the middle of your brain does 1 computation per second absolute maximum, and 1 per 10 seconds is more realistic. More to the outside (the vast majority of your brain) 1 per 100 seconds is a lot. And it slows down when you age.

This means brains must have a bloody good update rule. You just can't update a neural network in 1 billion operations per second, or 4e17 operations until you're 12, about 2 million training steps per neuron, or about half that assuming you sleep. You cannot get to the level of a 12 year old in 4e17 operations, because GPT-3 does more and while it's impressive, it doesn't have anything on a 12 year old.

Re: Predictive coding has been unified with backpropagation

#40

Post by eli_gottlieb » Mon, Apr 05, 2021, 8:02 PM UTC

tsmithe wrote: post #32

adamnemecek wrote: post #24

I think that this sort of forward backward thing is a very general idea. There’s a one to many relationship called the adjoint, and a many to one relationship called the norm. I wrote something about this here https://github.com/adamnemecek/adjoint

In fact, the compositional structure underlying that of predictive coding [0,1] is abstractly the same as that underlying backprop [2]. (Disclaimer: [0,1] are my own papers; I'm working on a more precise and extensive version of [1] right now!) [0] https://arxiv.org/abs/2006.01631 [1] https://arxiv.org/abs/2101.10483 [2] https://arxiv.org/abs/1711.10455

Hurry and publish before I have manuscripts ready applying these results.