NLP Wiki

User Tools

Site Tools

Trace:
ml:optimization_in_deep_learning

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
ml:optimization_in_deep_learning [2021/10/27 04:39] – jmflanigml:optimization_in_deep_learning [2025/03/25 00:49] (current) – [Effects on Optimization] jmflanig
Line 6: Line 6:
   * Weight normalization   * Weight normalization
     * Improves the conditioning of the optimization problem ([[https://arxiv.org/pdf/1602.07868.pdf|Salimans & Kingma 2016]])     * Improves the conditioning of the optimization problem ([[https://arxiv.org/pdf/1602.07868.pdf|Salimans & Kingma 2016]])
 +  * Lipschitz Constant
 +    * [[https://arxiv.org/pdf/2306.09338|Qi et al 2023 - Understanding Optimization of Deep Learning via Jacobian Matrix and Lipschitz Constant]] Talks about the effect of the Lipschitz constant on optimizing deep neural networks.
  
 ===== On Global Optimization of Neural Networks ===== ===== On Global Optimization of Neural Networks =====
Line 11: Line 13:
   * [[https://arxiv.org/pdf/1810.02054.pdf|Du et al 2018 - Gradient Descent Provably Optimizes Over-parameterized Neural Networks]] "We show that as long as m is large enough and no two inputs are parallel, randomly initialized gradient descent converges to a globally optimal solution at a linear convergence rate for the quadratic loss function, for an m hidden node shallow neural network with ReLU activation and n training data."   * [[https://arxiv.org/pdf/1810.02054.pdf|Du et al 2018 - Gradient Descent Provably Optimizes Over-parameterized Neural Networks]] "We show that as long as m is large enough and no two inputs are parallel, randomly initialized gradient descent converges to a globally optimal solution at a linear convergence rate for the quadratic loss function, for an m hidden node shallow neural network with ReLU activation and n training data."
   * [[https://arxiv.org/pdf/1910.00359.pdf|Goldblum et al 2020 - Truth or Backpropaganda? An Empirical Investigation of Deep Learning Theory]] Examines optimization issues in deep learning and their relation to ML theory   * [[https://arxiv.org/pdf/1910.00359.pdf|Goldblum et al 2020 - Truth or Backpropaganda? An Empirical Investigation of Deep Learning Theory]] Examines optimization issues in deep learning and their relation to ML theory
 +
 +===== Symmetries and Local Minima =====
 +  * [[http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.29.6931&rep=rep1&type=pdf|Sussman 1991 - Uniqueness of the weights for minimal feedforward nets with a given input-output map]] Also points out some discrete symmetries in neural networks: swapping any two neurons in the same layer, negating the input weights and output weights for any neuron (if using tahn activation), etc.
  
 ===== Properties of Neural Networks ===== ===== Properties of Neural Networks =====
Line 25: Line 30:
   * REINFORCE   * REINFORCE
   * [[https://www.aclweb.org/anthology/P18-1173.pdf|Peng et al 2018 - Backpropagating through Structured Argmax using a SPIGOT]]   * [[https://www.aclweb.org/anthology/P18-1173.pdf|Peng et al 2018 - Backpropagating through Structured Argmax using a SPIGOT]]
 +
 +===== Instabilities =====
 +
 +==== Instability of Fine-tuning ====
 +
 +  * [[https://arxiv.org/pdf/2006.04884.pdf|Mosbach 2020 - ON THE STABILITY OF FINE-TUNING BERT: MISCONCEPTIONS, EXPLANATIONS, AND STRONG BASELINES]] catastrophic forgetting and small size of the fine-tuning datasets fail to explain the fine-tuning instability, which is caused by optimization difficulties and differences in generalization.
 +
 +===== Implicit Regularization of SGD =====
 +See also Soheil Feizi's lecture [[https://www.youtube.com/watch?v=hGRssrXF-vI&list=PLHgjs9ncvHi80UCSlSvQe-TK_uOyDv_Jf&index=4|The implicit bias of gradient descent]].
 +  * [[https://www.jmlr.org/papers/volume19/18-188/18-188.pdf|Soudry et al 2018 - The Implicit Bias of Gradient Descent on Separable Data]]
 +  * [[https://arxiv.org/pdf/1806.01796.pdf|Nacson et al 2018 - Stochastic Gradient Descent on Separable Data: Exact Convergence with a Fixed Learning Rate]]
 +
 +===== Miscellaneous Topics =====
 +
 +==== Effect of Skip Connections ====
 +  * [[https://arxiv.org/pdf/1702.08591.pdf|Balduzzi et al 2017 - The Shattered Gradients Problem: If resnets are the answer, then what is the question?]]
 +  * [[https://arxiv.org/pdf/1701.09175.pdf|Orhan & Pitkow 2017 - Skip Connections Eliminate Singularities]]
 +
 +===== People =====
 +  * [[https://scholar.google.com/citations?user=AEBWEm8AAAAJ&hl=en|Daniel Soudry]]
  
 ===== Related Pages ===== ===== Related Pages =====
   * [[Optimization]]   * [[Optimization]]
  
ml/optimization_in_deep_learning.1635309549.txt.gz · Last modified: 2023/06/15 07:36 (external edit)
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki