Neuron-centric Hebbian Learning
2403.12076
0
0
📉
Abstract
One of the most striking capabilities behind the learning mechanisms of the brain is the adaptation, through structural and functional plasticity, of its synapses. While synapses have the fundamental role of transmitting information across the brain, several studies show that it is the neuron activations that produce changes on synapses. Yet, most plasticity models devised for artificial Neural Networks (NNs), e.g., the ABCD rule, focus on synapses, rather than neurons, therefore optimizing synaptic-specific Hebbian parameters. This approach, however, increases the complexity of the optimization process since each synapse is associated to multiple Hebbian parameters. To overcome this limitation, we propose a novel plasticity model, called Neuron-centric Hebbian Learning (NcHL), where optimization focuses on neuron- rather than synaptic-specific Hebbian parameters. Compared to the ABCD rule, NcHL reduces the parameters from $5W$ to $5N$, being $W$ and $N$ the number of weights and neurons, and usually $N ll W$. We also devise a ``weightless'' NcHL model, which requires less memory by approximating the weights based on a record of neuron activations. Our experiments on two robotic locomotion tasks reveal that NcHL performs comparably to the ABCD rule, despite using up to $sim97$ times less parameters, thus allowing for scalable plasticity
Create account to get full access
Overview
- The paper proposes a novel plasticity model called Neuron-centric Hebbian Learning (NcHL) that focuses on neuron-specific parameters rather than synaptic-specific parameters, reducing the number of parameters required.
- The paper also introduces a "weightless" NcHL model that further reduces memory requirements by approximating weights based on neuron activations.
- The authors evaluate NcHL on two robotic locomotion tasks and find it performs comparably to a previous model (ABCD rule) while using up to 97 times fewer parameters.
Plain English Explanation
The human brain has an amazing ability to adapt and change over time, called plasticity. This plasticity happens at the connections between neurons, called synapses, which transmit information across the brain.
Previous models of plasticity in artificial neural networks have focused on optimizing the properties of these synapses. However, this approach increases the complexity of the optimization process, as each synapse has multiple parameters to tune.
The researchers propose a new model called Neuron-centric Hebbian Learning (NcHL) that instead focuses on optimizing parameters specific to the neurons themselves, rather than the synapses. This reduces the total number of parameters that need to be tuned, making the optimization process more efficient.
They also develop a "weightless" version of NcHL that further reduces the memory requirements by approximating the synapse weights based on the neuron activations, rather than storing the weights explicitly.
When tested on robotic locomotion tasks, the NcHL model performed just as well as the previous synapse-focused model, but with dramatically fewer parameters - up to 97 times fewer. This suggests NcHL could enable more scalable plasticity in artificial neural networks.
Technical Explanation
The paper proposes a novel plasticity model called Neuron-centric Hebbian Learning (NcHL) that focuses on neuron-specific Hebbian parameters rather than synaptic-specific parameters, as used in previous models like the ABCD rule.
The ABCD rule optimizes multiple Hebbian parameters for each synapse, which increases the complexity of the optimization process. In contrast, NcHL reduces the number of parameters from 5W to 5N, where W is the number of weights and N is the number of neurons (typically N << W).
The authors also introduce a "weightless" version of NcHL that further reduces memory requirements. This model approximates the weights based on a record of neuron activations, rather than storing the weights explicitly.
The researchers evaluate NcHL on two robotic locomotion tasks and find it performs comparably to the ABCD rule, despite using up to ~97 times fewer parameters. This suggests NcHL could enable more efficient and hardware-aware training of neural networks by reducing the computational and memory requirements of plasticity.
Critical Analysis
The paper provides a compelling approach to reducing the complexity of plasticity models in artificial neural networks. By shifting the optimization focus from synapses to neurons, the authors are able to dramatically reduce the number of parameters that need to be tuned.
One potential limitation is that the paper only evaluates NcHL on two specific robotic locomotion tasks. It would be helpful to see how the model performs on a wider range of benchmarks and applications to better understand its general capabilities and limitations.
Additionally, the "weightless" version of NcHL, while promising in terms of memory savings, may introduce some accuracy tradeoffs by approximating the weights. The paper does not provide a detailed analysis of the performance impact of this approximation, which would be an important area for further research.
More broadly, the field of neural network plasticity is an active area of research, and it will be important to continue evaluating different approaches and their tradeoffs. Neuron-centric models like NcHL represent an intriguing direction, but there may be other novel approaches, such as spiking neural networks or node perturbation, that could also offer efficiency and scalability benefits.
Conclusion
The Neuron-centric Hebbian Learning (NcHL) model proposed in this paper represents an interesting step towards more efficient and scalable plasticity in artificial neural networks. By shifting the optimization focus from synapses to neurons, the authors are able to dramatically reduce the number of parameters required, while still maintaining comparable performance to a previous synapse-focused model.
The "weightless" variant of NcHL further reduces memory requirements, suggesting potential advantages for hardware-aware training and deployment. Overall, this work contributes to the ongoing research into novel plasticity mechanisms that could unlock more brain-inspired learning capabilities in artificial neural networks.
This summary was produced with help from an AI and may contain inaccuracies - check out the links to read the original source documents!
Related Papers
🤷
Unsupervised representation learning with Hebbian synaptic and structural plasticity in brain-like feedforward neural networks
Naresh Ravichandran, Anders Lansner, Pawel Herman
0
0
Neural networks that can capture key principles underlying brain computation offer exciting new opportunities for developing artificial intelligence and brain-like computing algorithms. Such networks remain biologically plausible while leveraging localized forms of synaptic learning rules and modular network architecture found in the neocortex. Compared to backprop-driven deep learning approches, they provide more suitable models for deploying on neuromorphic hardware and have greater potential for scalability on large-scale computing clusters. The development of such brain-like neural networks depends on having a learning procedure that can build effective internal representations from data. In this work, we introduce and evaluate a brain-like neural network model capable of unsupervised representation learning. It builds on the Bayesian Confidence Propagation Neural Network (BCPNN), which has earlier been implemented as abstract as well as biophyscially detailed recurrent attractor neural networks explaining various cortical associative memory phenomena. Here we developed a feedforward BCPNN model to perform representation learning by incorporating a range of brain-like attributes derived from neocortical circuits such as cortical columns, divisive normalization, Hebbian synaptic plasticity, structural plasticity, sparse activity, and sparse patchy connectivity. The model was tested on a diverse set of popular machine learning benchmarks: grayscale images (MNIST, Fashion-MNIST), RGB natural images (SVHN, CIFAR-10), QSAR (MUV, HIV), and malware detection (EMBER). The performance of the model when using a linear classifier to predict the class labels fared competitively with conventional multi-layer perceptrons and other state-of-the-art brain-like neural networks.
6/10/2024
Two Tales of Single-Phase Contrastive Hebbian Learning
Rasmus Kj{ae}r H{o}ier, Christopher Zach
0
0
The search for ``biologically plausible'' learning algorithms has converged on the idea of representing gradients as activity differences. However, most approaches require a high degree of synchronization (distinct phases during learning) and introduce substantial computational overhead, which raises doubts regarding their biological plausibility as well as their potential utility for neuromorphic computing. Furthermore, they commonly rely on applying infinitesimal perturbations (nudges) to output units, which is impractical in noisy environments. Recently it has been shown that by modelling artificial neurons as dyads with two oppositely nudged compartments, it is possible for a fully local learning algorithm named ``dual propagation'' to bridge the performance gap to backpropagation, without requiring separate learning phases or infinitesimal nudging. However, the algorithm has the drawback that its numerical stability relies on symmetric nudging, which may be restrictive in biological and analog implementations. In this work we first provide a solid foundation for the objective underlying the dual propagation method, which also reveals a surprising connection with adversarial robustness. Second, we demonstrate how dual propagation is related to a particular adjoint state method, which is stable regardless of asymmetric nudging.
6/26/2024
Evolving Self-Assembling Neural Networks: From Spontaneous Activity to Experience-Dependent Learning
Erwan Plantec, Joachin W. Pedersen, Milton L. Montero, Eleni Nisioti, Sebastian Risi
0
0
Biological neural networks are characterized by their high degree of plasticity, a core property that enables the remarkable adaptability of natural organisms. Importantly, this ability affects both the synaptic strength and the topology of the nervous systems. Artificial neural networks, on the other hand, have been mainly designed as static, fully connected structures that can be notoriously brittle in the face of changing environments and novel inputs. Building on previous works on Neural Developmental Programs (NDPs), we propose a class of self-organizing neural networks capable of synaptic and structural plasticity in an activity and reward-dependent manner which we call Lifelong Neural Developmental Program (LNDP). We present an instance of such a network built on the graph transformer architecture and propose a mechanism for pre-experience plasticity based on the spontaneous activity of sensory neurons. Our results demonstrate the ability of the model to learn from experiences in different control tasks starting from randomly connected or empty networks. We further show that structural plasticity is advantageous in environments necessitating fast adaptation or with non-stationary rewards.
6/17/2024
📉
Contribute to balance, wire in accordance: Emergence of backpropagation from a simple, bio-plausible neuroplasticity rule
Xinhao Fan, Shreesh P Mysore
0
0
Backpropagation (BP) has been pivotal in advancing machine learning and remains essential in computational applications and comparative studies of biological and artificial neural networks. Despite its widespread use, the implementation of BP in the brain remains elusive, and its biological plausibility is often questioned due to inherent issues such as the need for symmetry of weights between forward and backward connections, and the requirement of distinct forward and backward phases of computation. Here, we introduce a novel neuroplasticity rule that offers a potential mechanism for implementing BP in the brain. Similar in general form to the classical Hebbian rule, this rule is based on the core principles of maintaining the balance of excitatory and inhibitory inputs as well as on retrograde signaling, and operates over three progressively slower timescales: neural firing, retrograde signaling, and neural plasticity. We hypothesize that each neuron possesses an internal state, termed credit, in addition to its firing rate. After achieving equilibrium in firing rates, neurons receive credits based on their contribution to the E-I balance of postsynaptic neurons through retrograde signaling. As the network's credit distribution stabilizes, connections from those presynaptic neurons are strengthened that significantly contribute to the balance of postsynaptic neurons. We demonstrate mathematically that our learning rule precisely replicates BP in layered neural networks without any approximations. Simulations on artificial neural networks reveal that this rule induces varying community structures in networks, depending on the learning rate. This simple theoretical framework presents a biologically plausible implementation of BP, with testable assumptions and predictions that may be evaluated through biological experiments.
5/24/2024