Deep Learning Theory

Overview

Deep Learning Theory

整理了一些深度学习的理论相关内容,持续更新。

Overview

  1. Recent advances in deep learning theory 总结了目前深度学习理论研究的六个方向的一些结果,概述型,没做深入探讨(2021)。

    • 1.1 complexity and capacity-basedapproaches for analyzing the generalizability of deep learning;

    • 1.2 stochastic differential equations andtheir dynamic systems for modelling stochastic gradient descent and its variants, which characterizethe optimization and generalization of deep learning, partially inspired by Bayesian inference;

    • 1.3 thegeometrical structures of the loss landscape that drives the trajectories of the dynamic systems;

    • 1.4 theroles of over-parameterization of deep neural networks from both positive and negative perspectives;

    • 1.5 theoretical foundations of several special structures in network architectures;

    • 1.6 the increasinglyintensive concerns in ethics and security and their relationships with generalizability

Course

  1. Theory of Deep LearningTTIC,西北大学等组织的一系列课程和讲座,基础课程涉及DL的基础(符号化,简化后的数学问题和结论),信息论和学习,统计和计算,信息论,统计学习和强化学习(2020)。

  2. MathsDL-spring19,MathDL系列,18,19,20年均有。

    • 3.1 Geometry of Data

      • Euclidean Geometry: transportation metrics, CNNs , scattering.
      • Non-Euclidean Geometry: Graph Neural Networks.
      • Unsupervised Learning under Geometric Priors (Implicit vs explicit models, microcanonical, transportation metrics).
      • Applications and Open Problems: adversarial examples, graph inference, inverse problems.
    • 3.2 Geometry of Optimization and Generalization

      • Stochastic Optimization (Robbins & Munro, Convergence of SGD)
      • Stochastic Differential Equations (Fokker-Plank, Gradient Flow, Langevin + + Dynamics, links with SGD; open problems) Dynamics of Neural Network Optimization (Mean Field Models using Optimal Transport, Kernel Methods)
      • Landscape of Deep Learning Optimization (Tensor/Matrix factorization, Deep Nets; open problems).
      • Generalization in Deep Learning.
    • 3.3 Open qustions on Reinforcement Learning

Architecture

  1. Partial Differential Equations is All You Need for Generating Neural Architectures -- A Theory for Physical Artificial Intelligence Systems 将统计物理的反应扩散方程,量子力学中的薛定谔方程,傍轴光学中的亥姆霍兹方程统一整合到神经网络偏微分方程中(NPDE),利用有限元方法找到数值解,从离散过程中,构造了多层感知,卷积网络,和循环网络,并提供了优化方法L-BFGS等,主要是建立了经典物理模型和经典神经网络的联系(2021)。

Approximation

  1. NN Approximation Theory

Optimization

  1. SGD

  2. offconvex几个学术工作者维护的AI博客。

Geometry

  1. Optima transmission

Book

  1. Theory of Deep Learning(draft)Rong Ge 等(2019)。

  2. Spectral Learning on Matrices and TensorsMajid Janzamin等(2020)

  3. Deep Learning Architectures A Mathematical Approach(2020),你可以libgen获取,内容如其名字,大概包含:工业问题,DL基础(激活,结构,优化等),函数逼近,万有逼近,RELU等逼近新研究,函数表示,以及两大方向,信息角度,几何角度等相关知识,实际场景中的卷积,池化,循环,生成,随机网络等具体实用内容的数学化,另外附录集合论,测度论,概率论,泛函,实分析等基础知识。

  4. The Principles of Deep Learning Theory(2021)Daniel A. Roberts and Sho Yaida(mit),Beginning from a first-principles component-level picture of networks,本书解释了如何通过求解层到层迭代方程和非线性学习动力学来确定训练网络输出的准确描述。一个主要的结果是网络的预测是由近高斯分布描述的,网络的深度与宽度的纵横比控制着与无限宽度高斯描述的偏差。本书解释了这些有效深度网络如何从训练中学习非平凡的表示,并更广泛地分析非线性模型的表示学习机制。从近内核方法的角度来看,发现这些模型的预测对底层学习算法的依赖可以用一种简单而通用的方式来表达。为了获得这些结果,作者开发了表示组流(RG 流)的概念来表征信号通过网络的传播。通过将网络调整到临界状态,他们为梯度爆炸和消失问题提供了一个实用的解决方案。作者进一步解释了 RG 流如何导致近乎普遍的行为,从而可以将由不同激活函数构建的网络做类别划分。Altogether, they show that the depth-to-width ratio governs the effective model complexity of the ensemble of trained networks。利用信息理论,作者估计了模型性能最好的最佳深宽比,并证明了残差连接能将深度推向任意深度。利用以上理论工具,就可以更加细致的研究架构的归纳偏差,超参数,优化。

  5. Physics-based Deep Learning(2021)N. Thuerey, P. Holl,etc.github resources深度学习与物理学的联系。比如基于物理的损失函数,可微流体模拟,逆问题的求解,Navier-Stokes方程的前向模拟,Controlling Burgers’ Equation和强化学习的关系等。

Session

  1. Foundations of Deep Learning(2019),西蒙研究中心会议。
  2. Deep Learning Theory 4(2021, ICML)Claire Monteleoni主持...,深度学习理论会议4,包含论文和视频。
  3. Deep Learning Theory 5 (2021,ICML)MaYi主持...,深度学习理论会议5,包含论文和视频。

Others

  1. Theoretical issues in deep networks 表明指数型损失函数中存在隐式的正则化,其优化的结果和一般损失函数优化结果一致,优化收敛结果和梯度流的迹有关,目前还不能证明哪个结果最优(2020)。
  2. The Dawning of a New Erain Applied MathematicsWeinan E关于在DL的新处境下结合历史的工作范式给出的指导性总结(2021)。
  3. Mathematics of deep learning from Newton Institute
  4. DEEP NETWORKS FROM THE PRINCIPLE OF RATE REDUCTION,白盒神经网络。
  5. redunet_paper白盒神经网络代码。
  6. Theory of Deep Convolutional Neural Networks:Downsampling下采样的数学分析Ding-Xuan Zhou(2020)
  7. Theory of deep convolutional neural networks II: Spherical analysis还有III:radial functions 逼近,(2020)。不过这些工作到底如何,只是用数学转换了一下,理论上没做过多贡献,或者和实际结合没难么紧密,还不得而知。
  8. The Modern Mathematics of Deep Learning(2021)主要是deep laerning的数学分析描述,涉及的问题包括:超参数网络的通用能力,深度在深度模型中的核心作用,深度学习对维度灾难的克服,优化在非凸优化问题的成功,学习的表示特征的数学分析,为何深度模型在物理问题上有超常表现,模型架构中的哪些因素以何种方式影响不同任务的学习中的不同方面。
[CVPR2021 Oral] End-to-End Video Instance Segmentation with Transformers

VisTR: End-to-End Video Instance Segmentation with Transformers This is the official implementation of the VisTR paper: Installation We provide instru

Yuqing Wang 687 Jan 07, 2023
Optimizing Deeper Transformers on Small Datasets

DT-Fixup Optimizing Deeper Transformers on Small Datasets Paper published in ACL 2021: arXiv Detailed instructions to replicate our results in the pap

16 Nov 14, 2022
An extremely simple, intuitive, hardware-friendly, and well-performing network structure for LiDAR semantic segmentation on 2D range image. IROS21

FIDNet_SemanticKITTI Motivation Implementing complicated network modules with only one or two points improvement on hardware is tedious. So here we pr

YimingZhao 54 Dec 12, 2022
Code accompanying "Evolving spiking neuron cellular automata and networks to emulate in vitro neuronal activity," accepted to IEEE SSCI ICES 2021

Evolving-spiking-neuron-cellular-automata-and-networks-to-emulate-in-vitro-neuronal-activity Code accompanying "Evolving spiking neuron cellular autom

SOCRATES: Self-Organizing Computational substRATES 2 Dec 02, 2022
Implementation of BI-RADS-BERT & The Advantages of Section Tokenization.

BI-RADS BERT Implementation of BI-RADS-BERT & The Advantages of Section Tokenization. This implementation could be used on other radiology in house co

1 May 17, 2022
ZeroGen: Efficient Zero-shot Learning via Dataset Generation

ZEROGEN This repository contains the code for our paper “ZeroGen: Efficient Zero

Jiacheng Ye 31 Dec 30, 2022
Detection of drones using their thermal signatures from thermal camera through YOLO-V3 based CNN with modifications to encapsulate drone motion

Drone Detection using Thermal Signature This repository highlights the work for night-time drone detection using a using an Optris PI Lightweight ther

Chong Yu Quan 6 Dec 31, 2022
Code for technical report "An Improved Baseline for Sentence-level Relation Extraction".

RE_improved_baseline Code for technical report "An Improved Baseline for Sentence-level Relation Extraction". Requirements torch = 1.8.1 transformers

Wenxuan Zhou 74 Nov 29, 2022
Probabilistic Entity Representation Model for Reasoning over Knowledge Graphs

Implementation for the paper: Probabilistic Entity Representation Model for Reasoning over Knowledge Graphs, Nurendra Choudhary, Nikhil Rao, Sumeet Ka

Nurendra Choudhary 8 Nov 15, 2022
The GitHub repository for the paper: “Time Series is a Special Sequence: Forecasting with Sample Convolution and Interaction“.

SCINet This is the original PyTorch implementation of the following work: Time Series is a Special Sequence: Forecasting with Sample Convolution and I

386 Jan 01, 2023
Implementation of Memformer, a Memory-augmented Transformer, in Pytorch

Memformer - Pytorch Implementation of Memformer, a Memory-augmented Transformer, in Pytorch. It includes memory slots, which are updated with attentio

Phil Wang 60 Nov 06, 2022
Pytorch implementation of CVPR2021 paper "MUST-GAN: Multi-level Statistics Transfer for Self-driven Person Image Generation"

MUST-GAN Code | paper The Pytorch implementation of our CVPR2021 paper "MUST-GAN: Multi-level Statistics Transfer for Self-driven Person Image Generat

TianxiangMa 46 Dec 26, 2022
A code generator from ONNX to PyTorch code

onnx-pytorch Generating pytorch code from ONNX. Currently support onnx==1.9.0 and torch==1.8.1. Installation From PyPI pip install onnx-pytorch From

Wenhao Hu 94 Jan 06, 2023
Code for DeepXML: A Deep Extreme Multi-Label Learning Framework Applied to Short Text Documents

DeepXML Code for DeepXML: A Deep Extreme Multi-Label Learning Framework Applied to Short Text Documents Architectures and algorithms DeepXML supports

Extreme Classification 49 Nov 06, 2022
TART - A PyTorch implementation for Transition Matrix Representation of Trees with Transposed Convolutions

TART This project is a PyTorch implementation for Transition Matrix Representati

Lee Sael 2 Jan 19, 2022
CAMoE + Dual SoftMax Loss (DSL): Improving Video-Text Retrieval by Multi-Stream Corpus Alignment and Dual Softmax Loss

CAMoE + Dual SoftMax Loss (DSL): Improving Video-Text Retrieval by Multi-Stream Corpus Alignment and Dual Softmax Loss This is official implement of "

程星 87 Dec 24, 2022
"Neural Turing Machine" in Tensorflow

Neural Turing Machine in Tensorflow Tensorflow implementation of Neural Turing Machine. This implementation uses an LSTM controller. NTM models with m

Taehoon Kim 1k Dec 06, 2022
CLIPort: What and Where Pathways for Robotic Manipulation

CLIPort CLIPort: What and Where Pathways for Robotic Manipulation Mohit Shridhar, Lucas Manuelli, Dieter Fox CoRL 2021 CLIPort is an end-to-end imitat

246 Dec 11, 2022
Videocaptioning.pytorch - A simple implementation of video captioning

pytorch implementation of video captioning recommend installing pytorch and pyth

Yiyu Wang 2 Jan 01, 2022
PyTorch wrapper for Taichi data-oriented class

Stannum PyTorch wrapper for Taichi data-oriented class PRs are welcomed, please see TODOs. Usage from stannum import Tin import torch data_oriented =

86 Dec 23, 2022