AI Paper Atlas

Research map / Continual learning / Online and replay-based

Visualizing the PHATE of Neural Networks

Scott Gigante, Adam S Charles, Smita Krishnaswamy, Gal Mishne

NeurIPS 2019 · Continual learning › Online and replay-based · approach: Gradient projection and plasticity preservation · 17 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use the multislice kernel with PHATE instead of standard visualization methodsHidden-unit training traces on MNIST · MNISTintraslice neighborhood preservation: 0.05 -> 0.26 (k=10); 0.21 -> 0.45 (k=40)Improved
Use the multislice kernel with PHATE instead of the strongest comparatorHidden-unit training traces on MNIST · MNISTinterslice neighborhood preservation: 0.96 -> 0.95 (k=10); 0.94 -> 0.93 (k=40)Hurt
Use multislice PHATE instead of the strongest comparatorHidden-unit training traces on MNIST · MNISTloss correlation: 0.47 -> 0.81Improved
Use Naive Rehearsal instead of Adam2-layer MLP continual learning · Split MNIST task learningvalidation loss: 0.042 -> 0.047Hurt
Use Adagrad instead of Adam2-layer MLP continual learning · Split MNIST domain learningvalidation loss: 1.062 -> 0.462Improved
Use Adagrad instead of Adam2-layer MLP continual learning · Split MNIST class learningvalidation loss: 4.156 -> 1.884Improved
Preserve hidden-unit neighborhood structure across task switchesContinual-learning networks · Split MNIST continual learningvalidation loss: ARI and final validation loss have correlation rho = 0.94Improved
Apply dropout rather than vanilla training3-layer MLP · MNIST generalizationmemorization error: 0.04 -> -0.09Improved
Apply activity L1 regularization rather than vanilla training3-layer MLP · MNIST generalizationmemorization error: 0.04 -> 0.11Hurt
Apply activity L2 regularization rather than vanilla training3-layer MLP · MNIST generalizationmemorization error: 0.04 -> 0.12Hurt
Use training data alone to construct the visualizationM-PHATE · MNIST generalizationSpearman correlation between embedding variance and memorization: rho=-0.98 -> rho = -0.95No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy