Related papers: Neural Networks Learn Statistics of Increasing Complexity

Neural Networks Learn Statistics of Increasing Complexity

URL: http://arxiv.org/abs/2402.04362v3
Date: Wed, 09 Oct 2024 06:43:49 GMT
Title: Neural Networks Learn Statistics of Increasing Complexity
Authors: Nora Belrose, Quintin Pope, Lucia Quirke, Alex Mallen, Xiaoli Fern,
Abstract summary: Distributional simplicity bias (DSB) posits that neural networks learn low-order moments of the data distribution first. We show that networks automatically learn to perform well on maximum-entropy distributions whose low-order statistics match those of the training set early in training, then lose this ability later. We use optimal transport methods to surgically edit the low-order statistics of one class to match those of another, and show that early-training networks treat the edited samples as if they were drawn from the target class.
Score: 2.1004767452202637
License:
Abstract: The distributional simplicity bias (DSB) posits that neural networks learn low-order moments of the data distribution first, before moving on to higher-order correlations. In this work, we present compelling new evidence for the DSB by showing that networks automatically learn to perform well on maximum-entropy distributions whose low-order statistics match those of the training set early in training, then lose this ability later. We also extend the DSB to discrete domains by proving an equivalence between token $n$-gram frequencies and the moments of embedding vectors, and by finding empirical evidence for the bias in LLMs. Finally we use optimal transport methods to surgically edit the low-order statistics of one class to match those of another, and show that early-training networks treat the edited samples as if they were drawn from the target class. Code is available at https://github.com/EleutherAI/features-across-time.

Related papers

Sebra: Debiasing Through Self-Guided Bias Ranking [54.09529903433859]
Ranking samples by fine-grained estimates of spuriosity has recently been shown to significantly benefit bias mitigation. We propose a debiasing framework based on our novel ulSelf-Guided ulBias ulRanking (emphSebra) Sebra mitigates biases via an automatic ranking of data points by spuriosity within their respective classes.
arXiv Detail & Related papers (2025-01-30T11:31:38Z)
Benign Overfitting and Grokking in ReLU Networks for XOR Cluster Data [42.870635753205185]
Neural networks trained by gradient descent (GD) have exhibited a number of surprising generalization behaviors. We show that both of these phenomena provably occur in two-layer ReLU networks trained by GD on XOR cluster data. At a later training step, the network achieves near-optimal test accuracy while still fitting the random labels in the training data, exhibiting a "grokking" phenomenon.
arXiv Detail & Related papers (2023-10-04T02:50:34Z)
Neural networks trained with SGD learn distributions of increasing complexity [78.30235086565388]
We show that neural networks trained using gradient descent initially classify their inputs using lower-order input statistics. We then exploit higher-order statistics only later during training. We discuss the relation of DSB to other simplicity biases and consider its implications for the principle of universality in learning.
arXiv Detail & Related papers (2022-11-21T15:27:22Z)
Learning an Invertible Output Mapping Can Mitigate Simplicity Bias in Neural Networks [66.76034024335833]
We investigate why diverse/ complex features are learned by the backbone, and their brittleness is due to the linear classification head relying primarily on the simplest features. We propose Feature Reconstruction Regularizer (FRR) to ensure that the learned features can be reconstructed back from the logits. We demonstrate up to 15% gains in OOD accuracy on the recently introduced semi-synthetic datasets with extreme distribution shifts.
arXiv Detail & Related papers (2022-10-04T04:01:15Z)
On the Implicit Bias of Gradient Descent for Temporal Extrapolation [32.93066466540839]
Common practice when using recurrent neural networks (RNNs) is to apply a model to sequences longer than those seen in training. We show that even with infinite training data, there exist RNN models that interpolate perfectly. We then show that if gradient descent is used for training, learning will converge to perfect extrapolation.
arXiv Detail & Related papers (2022-02-09T06:28:37Z)
Transformers Can Do Bayesian Inference [56.99390658880008]
We present Prior-Data Fitted Networks (PFNs) PFNs leverage in-context learning in large-scale machine learning techniques to approximate a large set of posteriors. We demonstrate that PFNs can near-perfectly mimic Gaussian processes and also enable efficient Bayesian inference for intractable problems.
arXiv Detail & Related papers (2021-12-20T13:07:39Z)
The Lottery Ticket Hypothesis for Pre-trained BERT Networks [137.99328302234338]
In natural language processing (NLP), enormous pre-trained models like BERT have become the standard starting point for training. In parallel, work on the lottery ticket hypothesis has shown that models for NLP and computer vision contain smaller matchingworks capable of training in isolation to full accuracy. We combine these observations to assess whether such trainable, transferrableworks exist in pre-trained BERT models.
arXiv Detail & Related papers (2020-07-23T19:35:39Z)
Learning from Failure: Training Debiased Classifier from Biased Classifier [76.52804102765931]
We show that neural networks learn to rely on spurious correlation only when it is "easier" to learn than the desired knowledge. We propose a failure-based debiasing scheme by training a pair of neural networks simultaneously. Our method significantly improves the training of the network against various types of biases in both synthetic and real-world datasets.
arXiv Detail & Related papers (2020-07-06T07:20:29Z)

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