There is a lot of interesting research into predictive coding as an alternative means to solve the credit assignment problem that might be a more plausible model of what happens in the brain.
I really liked this paper that showed using a predictive coding learning rule leads to the exact same gradients as backprop in arbitrary networks:
I wonder if you could take a traditional backprop trained LLM and apply this approach to finetuning it (presumably needs less memory and compute?). It could be another entry in the spectrum between LORA and full fine tuning.
Nothing prevents interleaving inference and gradient descent / RMAD / backpropagation from being interleaved.
Is a man a father or a son? It's a false dilemma, it can be both.
Is a training step pretraining or finetuning? It's the same mathematical operation, with the only difference the intention or reason of applying the training step.
Oh wow the theoretical implications in neuroscience exite me here - is this a potential model of Fristons Markov Blanket concept
“ Probably the most ambitious and all-encompassing version of the ‘Bayesian turn’ in cognitive science is
the free energy principle (FEP). The FEP is a mathematical framework, developed by Karl Friston and
colleagues (Friston, Kilner, and Harrison 2006; Friston et al. 2010; Friston 2010; Friston et al. 2017a;
Friston 2019), which specifies an objective function that any self-organizing system needs to minimize in
order to ensure adaptive exchanges with its environment. One major appeal of the FEP is that it aims for
(and seems to deliver) an unprecedented integration of the life sciences (including psychology,
neuroscience, and theoretical biology). The difference between the FEP and earlier inferential theories
(e.g., Gregory 1980, Grossberg 1980, Rao and Ballard 1999, Lee and Mumford 2003) is that not only
perceptual processes, but also other cognitive functions such as learning, attention, and action planning
can be subsumed under one single principle: the minimization of free energy through the process of active
inference (Friston 2010; Friston et al. 2017). ”
Isn't the FEP basically just loss minimization over KL-divergence? In other words, it's the same thing we already do with ML and already have been doing for years? I've never understood where this differs to the status quo, or why this isn't just a relabelling of techniques/concepts. Although I didn't look too deeply.
It might be beneficial while not being optimal on its own.
The obvious example is if it has different behaviour around local minima, it could be an altenate pathway out.
I have often wondered if doing training with radically different aproaches for the first few iterarions would avoid any method specific artifacts before the weights had time to denoise.
If we do find a fruitful new technique (and maybe this is one), I feel like there's a chance it's overlooked because we evaluate it on the wrong metrics, and that it needs to be scaled up considerably before it can be judged fairly. Techniques that can train a model to generalize better up-front will look weaker on benchmarks until they're scaled and tested against underlying patterns that transfer across unseen samples, rather than just predicting an exact outcome precisely. It may be the case that the optimal system has an incredibly slow, painful loss curve at first, and then wildly darts down at some unknown threshold of scale.
Another example is that we're building architectures that are incredible at short-term memory - if we consider the context window like the brains short term memory. A human can never have the level of awareness over such a large amount of information. But humans leave chains of fuzzy pointers that somehow associatively bind to future expectations and the behavioral techniques that allow us to trace back when and where we need to, quite efficiently. An LLM works very differently and is poor at getting a "feeling" that it's missing information. For us it's as if, in real time, we mould our neural network such that the future stimuli will automatically trigger these pathways without additional processing.
All this is to say: We've developed frameworks that measure performance of existing LLMs relative to LLMs, but not relative to different paradigms that might actually end up having greater intelligence or flexibility at scale, or interesting properties and behaviors that work very differently to LLMs.
I also wonder if our short-term memory being weak is actually a forcing function (bottleneck) for unlocking certain capabilities in a general system. We build complex sophisticated machines to process, store, manage and maintain information for us. Why bother learning to do that if you can just remember everything?
Personally I think the dirty secret of the brain is that a lot of things are hard coded. And many things that we need to learn are also hard coded except that some parameters need to be tuned.
If we puke, the brain will not do general aversive learning, it will learn to avoid specifically the last thing eaten, because it instinctively knows about food poisoning.
Imprinting is absolutely fascinating. Some newborn animals will run a very simple pattern detector like looking for a red dot or something and use that to bootstrap their conception of their parent.
For fully general learning I have a hunch that it can be done using local history plus a semi-global reward scalar (global neurotransmittor levels).
There is a lot of interesting research into predictive coding as an alternative means to solve the credit assignment problem that might be a more plausible model of what happens in the brain.
I really liked this paper that showed using a predictive coding learning rule leads to the exact same gradients as backprop in arbitrary networks:
Predictive Coding Approximates Backprop Along Arbitrary Computation Graphs https://direct.mit.edu/neco/article/34/6/1329/110646/Predict...
I wonder if you could take a traditional backprop trained LLM and apply this approach to finetuning it (presumably needs less memory and compute?). It could be another entry in the spectrum between LORA and full fine tuning.
Could this relate to continual learning? It lets you update without pausing the entire system.
Nothing prevents interleaving inference and gradient descent / RMAD / backpropagation from being interleaved.
Is a man a father or a son? It's a false dilemma, it can be both.
Is a training step pretraining or finetuning? It's the same mathematical operation, with the only difference the intention or reason of applying the training step.
Oh wow the theoretical implications in neuroscience exite me here - is this a potential model of Fristons Markov Blanket concept
“ Probably the most ambitious and all-encompassing version of the ‘Bayesian turn’ in cognitive science is the free energy principle (FEP). The FEP is a mathematical framework, developed by Karl Friston and colleagues (Friston, Kilner, and Harrison 2006; Friston et al. 2010; Friston 2010; Friston et al. 2017a; Friston 2019), which specifies an objective function that any self-organizing system needs to minimize in order to ensure adaptive exchanges with its environment. One major appeal of the FEP is that it aims for (and seems to deliver) an unprecedented integration of the life sciences (including psychology, neuroscience, and theoretical biology). The difference between the FEP and earlier inferential theories (e.g., Gregory 1980, Grossberg 1980, Rao and Ballard 1999, Lee and Mumford 2003) is that not only perceptual processes, but also other cognitive functions such as learning, attention, and action planning can be subsumed under one single principle: the minimization of free energy through the process of active inference (Friston 2010; Friston et al. 2017). ”
Isn't the FEP basically just loss minimization over KL-divergence? In other words, it's the same thing we already do with ML and already have been doing for years? I've never understood where this differs to the status quo, or why this isn't just a relabelling of techniques/concepts. Although I didn't look too deeply.
Rolls right off the tongue
~85% accuracy on MNIST. Sigh.
How does it do on CIFAR-10, or even better, ImageNet?
Interesting research, not sure it's a backprop alternative.
===
EDIT: accuracy on MNIST is not ~90%. It's ~85%.
Their image classification benchmarks include both: https://pub.sakana.ai/pc-alm/assets/figures/benchmark_accura...
~74% on CIFAR-10. Still a far cry from backprop.
I didn't see ImageNet. TinyImageNet is something else.
It might be beneficial while not being optimal on its own.
The obvious example is if it has different behaviour around local minima, it could be an altenate pathway out.
I have often wondered if doing training with radically different aproaches for the first few iterarions would avoid any method specific artifacts before the weights had time to denoise.
You can probably distribute training more easily too
If we do find a fruitful new technique (and maybe this is one), I feel like there's a chance it's overlooked because we evaluate it on the wrong metrics, and that it needs to be scaled up considerably before it can be judged fairly. Techniques that can train a model to generalize better up-front will look weaker on benchmarks until they're scaled and tested against underlying patterns that transfer across unseen samples, rather than just predicting an exact outcome precisely. It may be the case that the optimal system has an incredibly slow, painful loss curve at first, and then wildly darts down at some unknown threshold of scale.
Another example is that we're building architectures that are incredible at short-term memory - if we consider the context window like the brains short term memory. A human can never have the level of awareness over such a large amount of information. But humans leave chains of fuzzy pointers that somehow associatively bind to future expectations and the behavioral techniques that allow us to trace back when and where we need to, quite efficiently. An LLM works very differently and is poor at getting a "feeling" that it's missing information. For us it's as if, in real time, we mould our neural network such that the future stimuli will automatically trigger these pathways without additional processing.
All this is to say: We've developed frameworks that measure performance of existing LLMs relative to LLMs, but not relative to different paradigms that might actually end up having greater intelligence or flexibility at scale, or interesting properties and behaviors that work very differently to LLMs.
I also wonder if our short-term memory being weak is actually a forcing function (bottleneck) for unlocking certain capabilities in a general system. We build complex sophisticated machines to process, store, manage and maintain information for us. Why bother learning to do that if you can just remember everything?
They state replacing backprop is not their goal.
Their goal is to understand how distributed systems which cannot do backprop (the brain) can still do learning.
Personally I think the dirty secret of the brain is that a lot of things are hard coded. And many things that we need to learn are also hard coded except that some parameters need to be tuned.
If we puke, the brain will not do general aversive learning, it will learn to avoid specifically the last thing eaten, because it instinctively knows about food poisoning.
Imprinting is absolutely fascinating. Some newborn animals will run a very simple pattern detector like looking for a red dot or something and use that to bootstrap their conception of their parent.
For fully general learning I have a hunch that it can be done using local history plus a semi-global reward scalar (global neurotransmittor levels).
New paper by Sakana.ai [1]
[1]: https://arxiv.org/abs/2605.31022