A curated list of papers of interesting empirical study and insight on deep learning. Continually updating...

Code for Arxiv Double Descent Demystified: Identifying, Interpreting & Ablating the Sources of a Deep Learning Puzzle

MDL Complexity computations and experiments from the paper "Revisiting complexity and the bias-variance tradeoff".

Double Descent results for FCNNs on MNIST, extended by Label Noise (Reconciling Modern Machine-Learning Practice and the Classical Bias–Variance Trade-Off) [Python/PyTorch]..

Explore the double-descent phenomena in the context of system identification. Companion code to the paper (https://arxiv.org/abs/2012.06341):

[ICIAP23] Sparse Double Descent in Vision Transformers: real or phantom threat?

This repository is the official implementation of "Optimization Variance: Delve into the Epoch-Wise Double Descent of DNNs"

[AAAI24] DSD2: Can We Dodge Sparse Double Descent and Compress the Neural Network Worry-Free?

Double descent experiments/repros on classical ML models and deep neural nets

Assignments of my CST Part II Deep Neural Networks unit

A Review of Preetum Nakkiran's "More Data Can Hurt for Linear Regression: Sample-wise Double Descent"

This project outlines 4 experiments to explore the effects of several settings on the bias-variance tradeoff curve

Implementation of the double descent Deep Learning phenomenon from the article Grokking: Generalization beyond overfitting.

ICLR 2022: Phenomenology of Double Descent in Finite-width Neural Networks

Interpolating Neural Networks in Asset Pricing Data. Supports Distributed Training in TensorFlow.

Toy dataset to study double descent optimization patterns in machine learning.

교과서의 편향-분산 U자 곡선이 깨지는 두 현상(더블 디센트·그로킹)을 브라우저 안에서 실제 수치 시뮬레이션으로 재현하는 단일 HTML 인터랙티브 실험실 (AI기초 심화 탐구)

Do sparse, one-class purchase data exhibit double descent? A study of NMF, weighted MF (iALS), and Logistic Matrix Factorization on real e-commerce transactions. Sapienza SML project.

This assignment repository demonstrates the existance of deep double descent phenomena for the deep learning models usually encorporated with a clinical context. For this, a famous ECG dataset, PTBXL has been employed.

Block-Term Operator Theory: why block-term rank-(L,L,1) neural operators generalize better than CP / Tucker / TT at matched capacity, not by more expressivity but as a tighter inductive bias. A least-squares generalization separation Theta((RL - mu_band) K / n), a complete variance-ordering theorem across all four tensor formats, an adaptive for...