Not All Explanations for Deep Learning Phenomena Are Equally Valuable
Not on the map. Why?
Abstract (arXiv)
Developing a better understanding of surprising or counterintuitive phenomena has constituted a significant portion of deep learning research in recent years. These include double descent, grokking, and the lottery ticket hypothesis -- among many others. Works in this area often develop ad hoc hypotheses attempting to explain these observed phenomena on an isolated, case-by-case basis. This position paper asserts that, in many prominent cases, there is little evidence to suggest that these phenomena appear in real-world applications and these efforts may be inefficient in driving progress in the broader field. Consequently, we argue against viewing them as isolated puzzles that require bespoke resolutions or explanations. However, despite this, we suggest that deep learning phenomena do still offer research value by providing unique settings in which we can refine our broad explanatory theories of more general deep learning principles. This position is reinforced by analyzing the research outcomes of several prominent examples of these phenomena from the recent literature. We revisit the current norms in the research community in approaching these problems and propose practical recommendations for future research, aiming to ensure that progress on deep learning phenomena is well aligned with the ultimate pragmatic goal of progress in the broader field of deep learning.
Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Builds on 53 of its 80 references that are in the map. First posted on arXiv 29 Jun 2025.
Builds on
53 of 80 references.
References · 53
- 2020Scaling Laws for Neural Language Modelscited by 6,586
- 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,926
- 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,673
- 2018The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networkscited by 2,262
- 2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,298
- 2021BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-modelscited by 1,245
- 2021Scaling Vision Transformerscited by 999
- 2018Reconciling modern machine learning practice and the bias-variance trade-offcited by 927
- 2023Progress measures for grokking via mechanistic interpretabilitycited by 839
- 2021Going deeper with Image Transformerscited by 764
- 2022Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasetscited by 629
- 2018Rethinking the Value of Network Pruningcited by 599
38 more
- 2019Benign Overfitting in Linear Regressioncited by 549
- 2021Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networkscited by 527
- 2023Are Emergent Abilities of Large Language Models a Mirage?cited by 523
- 2019Surprises in High-Dimensional Ridgeless Least Squares Interpolationcited by 513
- 2023Faith and Fate: Limits of Transformers on Compositionalitycited by 503
- 2020Underspecification Presents Challenges for Credibility in Modern Machine Learningcited by 496
- 2019Does Learning Require Memorization? A Short Tale about a Long Tailcited by 460
- 2020Improving Reproducibility in Machine Learning Research (A Report from the NeurIPS 2019 Reproducibility Program)cited by 322
- 2023Challenges and Applications of Large Language Modelscited by 254
- 2022Towards Understanding Grokking: An Effective Theory of Representation Learningcited by 230
- 2019Deconstructing Lottery Tickets: Zeros, Signs, and the Supermaskcited by 207
- 2020Proving the Lottery Ticket Hypothesis: Pruning is All You Needcited by 201
- 2020Pruning Neural Networks at Initialization: Why are We Missing the Mark?cited by 182
- 2021The Principles of Deep Learning Theorycited by 173
- 2021The Values Encoded in Machine Learning Researchcited by 172
- 2020Optimal Regularization Can Mitigate Double Descentcited by 109
- 2023Grokking as the Transition from Lazy to Rich Training Dynamicscited by 108
- 2023Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokkingcited by 80
- 2020Understanding Double Descent Requires a Fine-Grained Bias-Variance Decompositioncited by 76
- 2023Grokking of Hierarchical Structure in Vanilla Transformerscited by 53
- 2021Sanity Checks for Lottery Tickets: Does Your Winning Ticket Really Win the Jackpot?cited by 52
- 2025Grokking at the Edge of Numerical Stabilitycited by 46
- 2020A Brief Prehistory of Double Descentcited by 44
- 2020Early Stopping in Deep Networks: Double Descent and How to Eliminate itcited by 41
- 2025Emergent Abilities in Large Language Models: A Surveycited by 37
- 2023Learning the greatest common divisor: explaining transformer predictionscited by 33
- 2023A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learningcited by 33
- 2022Coarsening the Granularity: Towards Structurally Sparse Lottery Ticketscited by 27
- 2020Scaling Down Deep Learning with MNIST-1Dcited by 25
- 2024A Survey of Lottery Ticket Hypothesiscited by 24
- 2023Joint Training of Deep Ensembles Fails Due to Learner Collusioncited by 23
- 2024Emergent Equivariance in Deep Ensemblescited by 18
- 2019The Scientific Method in the Science of Machine Learningcited by 15
- 2024Why do Random Forests Work? Understanding Tree Ensembles as Self-Regularizing Adaptive Smootherscited by 15
- 2019HARK Side of Deep Learning -- From Grad Student Descent to Automated Machine Learningcited by 13
- 2023SoK: Pragmatic Assessment of Machine Learning for Network Intrusion Detectioncited by 13
- 2023Grokking Beyond Neural Networks: An Empirical Exploration with Model Complexitycited by 13
- 2023Pre-registration for Predictive Modelingcited by 13
50 most cited of 53 All 53 (JSON)
Built on by
Citing papers per year: 2026: 3.
3 papers.