Abstract The brain processes information through multiple layers of neurons. This deep architecture is representationally powerful, but complicates learning because it is difficult to identify the responsible neurons when a mistake is made. In machine learning, the backpropagation algorithm assigns blame by multiplying error signals with all the synaptic weights on each neuron’s axon and further downstream. However, this involves a precise, symmetric backward connectivity pattern, which is thought to be impossible in the brain. Here we demonstrate that this strong architectural constraint is not required for effective error propagation. We present a surprisingly simple mechanism that assigns blame by multiplying errors by even random synaptic weights. This mechanism can transmit teaching signals across multiple layers of neurons and performs as effectively as backpropagation on a variety of tasks. Our results help reopen questions about how the brain could use error signals and dispel long-held assumptions about algorithmic constraints on learning.
Metadata: Crossref. Abstracts retain their original authors’ rights.
In the conversation
Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels
Neat. That thousand brains site looks right up my alley. If you haven't seen it, maybe check this out: https://www.nature.com/articles/ncomms13276
The idea is that the brain uses what the authors refer to as "feedback alignment" rather than backprop. Even if it turns out not to be literally true of the brain, the idea is interesting for AI.
I also love the idea of grafting on the memory banks. It reminds me of early work on DNC's (Differentiable Neural Computer's). I tried to franken-bolt a DNC onto an LLM a few years back and mostly just earned myself headaches. :)
It's fun to see all the wild and wacky stuff other folks like myself are tinkering with in the lab.
> For example, let's say instead of gradient descent you want to do a Newton descent. Then maybe there's a better way to compute the needed weight updates besides backprop?
IIRC, feedback alignment [1] approximates Gauss-Newton minimization. So there is an easier way, that is potentially biologically more plausible, though not necessarily a better way.
[1] https://www.nature.com/articles/ncomms13276#Sec20
> recurrent projections would need to (1) form a 3rd party connection at every synapse and (2) know which synapses were to blame for the error
The most striking result in this regard is that one can get backprop with random backprojections. It works regardless because, on average, the error vector will be less than 90° from the true error vector (which is good enough for hillclimbing) and the dynamics play out in a way that the learned weights adjust to the random projections:
https://www.nature.com/articles/ncomms13276
That being said, it does seem to be the case that the brain simply memorizes an awful lot which must work by a different mechanism besides backprop…
Read the full comment
because backprop cannot do one-shot learning. I think one-shot learning is how the brain gets past large discontinuities in the model fitness landscape: It can learn linguistic, logical rules, fragments of general computations and facts that are discovered by exploration (which includes learning about whether it was Bob or Mike) and passed on culturally. The brain basically outsources the problem of tunneling through large discontinuities, to cultural/individual trial-and-error and episodic memory. The greatest consequence is that these bodies of knowledge can concern the improvement of the organization of knowledge itself, resulting in a positive feedback loop in model fitness (especially science/Bayesian updating). Though such bodies of knowledge evolve respecting a learnability-by-hillclimbing soft constraint, which implies they often form a neat latent space where similar codes are organized to belong to similar meanings/representations, as this can easily be learned by stochastic hillclimbing (repetition) because each time the brain processes related information, it is nudged towards the latent space that is meant to be learned. Many parts of the world happen to be learnable in this way because everything is kinda smooth and continuous. Small causes tend to have small effects as everything consists of a myriad of small particles that affect each other in smooth ways if you squint at them. Though obviously not everything can be learned this way (implying large discontinuities) which is where a brute memorization based on reward and punishment comes in handy.
Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task.
Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small difference between training and test performance.
Fast compilation is important when compilation occurs at runtime, such as query compilers in modern database systems and WebAssembly virtual machines in modern browsers.
In honor of Alan Turing's hundredth birthday, I unwisely set out some thoughts about one of Turing's obsessions throughout his life, the question of physics and free will.