Sparse Inter-Layer Dependencies of Transformer FFN Neurons

• URL: http://arxiv.org/abs/2607.11990v1
• Date: Mon, 13 Jul 2026 15:04:50 GMT
• Title: Sparse Inter-Layer Dependencies of Transformer FFN Neurons
• Abstract summary: We study whether the activation of an FFN neuron can be explained by a sparse set of preceding neuron activations and attention outputs.<n>We introduce a training-free attribution method that estimates the relative influence of upstream neurons and attention outputs on a target neuron's activation.
• Score: 32.00544060412007
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Feedforward network (FFN) blocks account for a large fraction of the parameters and computation in Transformer architectures, yet their internal structure remains difficult to interpret due to the additive superposition induced by the residual stream. We examine whether the activation of an FFN neuron can be explained by a sparse set of preceding neuron activations and attention outputs. We introduce a training-free attribution method that estimates the relative influence of upstream neurons and attention outputs on a target neuron's activation. Empirically, across models and layers, we find that small subsets of preceding activations and attention outputs suffice to preserve neuron activations with high fidelity when all remaining inputs are masked with their average values. Effective sparsity is even greater when accounting for the inherent activation sparsity of upstream layers. Moreover, applying the neuron-specific masks in all layers simultaneously, such that the induced deviations propagate through the network, leaves model perplexity largely unchanged at moderate sparsity levels. These results demonstrate that, despite dense parameterization, FFNs exhibit sparse and structured inter-layer dependencies at the neuron level. Our method provides a practical, scalable tool for circuit-level interpretability and identifies candidate sparse pathways with potential implications for efficient inference.

Related papers

• General Self-Prediction Enhancement for Spiking Neurons [71.01912385372577]
  Spiking Neural Networks (SNNs) are highly energy-efficient due to event-driven, sparse computation, but their training is challenged by spike non-differentiability and trade-offs among performance, efficiency, and biological plausibility.<n>We propose a self-prediction enhanced spiking neuron method that generates an internal prediction current from its input-output history to modulate membrane potential.<n>This design offers dual advantages, it creates a continuous gradient path that alleviates vanishing gradients and boosts training stability and accuracy, while also aligning with biological principles, which resembles distal dendritic modulation and error-driven synaptic plasticity.
  arXiv  Detail & Related papers  (2026-01-29T15:08:48Z)
• Training Deep Normalization-Free Spiking Neural Networks with Lateral Inhibition [52.59263087086756]
  Training deep neural networks (SNNs) has critically depended on explicit normalization schemes, such as batch normalization.<n>We propose a normalization-free learning framework that incorporates lateral inhibition inspired by cortical circuits.<n>We show that our framework enables stable training of deep SNNs with biological realism and achieves competitive performance without resorting to explicit normalizations.
  arXiv  Detail & Related papers  (2025-09-27T11:11:30Z)
• Understanding Artificial Neural Network's Behavior from Neuron Activation Perspective [8.251799609350725]
  This paper explores the intricate behavior of deep neural networks (DNNs) through the lens of neuron activation dynamics.<n>We propose a probabilistic framework that can analyze models' neuron activation patterns as a process.
  arXiv  Detail & Related papers  (2024-12-24T01:01:06Z)
• Towards Utilising a Range of Neural Activations for Comprehending Representational Associations [0.6554326244334868]
  We show that an approach to label intermediate representations in deep neural networks fails to capture valuable information about their behaviour. We hypothesise that non-extremal level activations contain complex information worth investigating. We use our findings to develop a method to curate data from mid-range logit samples for retraining to mitigate spurious correlations.
  arXiv  Detail & Related papers  (2024-11-15T07:54:14Z)
• Confidence Regulation Neurons in Language Models [91.90337752432075]
  This study investigates the mechanisms by which large language models represent and regulate uncertainty in next-token predictions. Entropy neurons are characterized by an unusually high weight norm and influence the final layer normalization (LayerNorm) scale to effectively scale down the logits. token frequency neurons, which we describe here for the first time, boost or suppress each token's logit proportionally to its log frequency, thereby shifting the output distribution towards or away from the unigram distribution.
  arXiv  Detail & Related papers  (2024-06-24T01:31:03Z)
• ProAct: Progressive Training for Hybrid Clipped Activation Function to Enhance Resilience of DNNs [0.4660328753262075]
  State-of-the-art methods offer either neuron-wise or layer-wise clipping activation functions. Layer-wise clipped activation functions cannot preserve DNNs resilience at high bit error rates. We propose a hybrid clipped activation function that integrates neuron-wise and layer-by-layer methods.
  arXiv  Detail & Related papers  (2024-06-10T14:31:38Z)
• Fully Spiking Actor Network with Intra-layer Connections for Reinforcement Learning [51.386945803485084]
  We focus on the task where the agent needs to learn multi-dimensional deterministic policies to control. Most existing spike-based RL methods take the firing rate as the output of SNNs, and convert it to represent continuous action space (i.e., the deterministic policy) through a fully-connected layer. To develop a fully spiking actor network without any floating-point matrix operations, we draw inspiration from the non-spiking interneurons found in insects.
  arXiv  Detail & Related papers  (2024-01-09T07:31:34Z)
• Globally Optimal Training of Neural Networks with Threshold Activation Functions [63.03759813952481]
  We study weight decay regularized training problems of deep neural networks with threshold activations. We derive a simplified convex optimization formulation when the dataset can be shattered at a certain layer of the network.
  arXiv  Detail & Related papers  (2023-03-06T18:59:13Z)
• And/or trade-off in artificial neurons: impact on adversarial robustness [91.3755431537592]
  Presence of sufficient number of OR-like neurons in a network can lead to classification brittleness and increased vulnerability to adversarial attacks. We define AND-like neurons and propose measures to increase their proportion in the network. Experimental results on the MNIST dataset suggest that our approach holds promise as a direction for further exploration.
  arXiv  Detail & Related papers  (2021-02-15T08:19:05Z)
• Activation function dependence of the storage capacity of treelike neural networks [16.244541005112747]
  nonlinear activation functions have been proposed for use in artificial neural networks. We study how activation functions affect the storage capacity of treelike two-layer networks.
  arXiv  Detail & Related papers  (2020-07-21T23:51:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.