ActUp: Analyzing and Consolidating tSNE and UMAP

Authors: Andrew Draganov, Jakob Jørgensen, Katrine Scheel, Davide Mottin, Ira Assent, Tyrus Berry, Cigdem Aslay

IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally validate that GDR can simulate both methods through a thorough quantitative and qualitative evaluation across many datasets and settings.
Researcher Affiliation Academia 1Aarhus University 2George Mason University {draganovandrew, jakobrj, scheel, davide, ira, cigdem}@cs.au.dk, tberry@gmu.edu
Pseudocode No The paper does not contain any sections or figures explicitly labeled as "Pseudocode" or "Algorithm".
Open Source Code Yes We release improved versions of t SNE, UMAP, and GDR that are fully plug-andplay with the traditional libraries.
Open Datasets Yes The paper uses well-known public datasets such as MNIST [Deng, 2012], Fashion MNIST [Xiao et al., 2017], Coil-100 [NENE, 1996], Single Cell [Tasic et al., 2018], and Cifar-10 [Krizhevsky, 2009].
Dataset Splits No The paper mentions "training" and metrics like "k NN accuracy" and "V-measure" but does not specify train/validation/test splits, percentages, or absolute sample counts for data partitioning.
Hardware Specification No The paper mentions running computations "on a GPU" but does not specify any particular GPU model (e.g., NVIDIA A100), CPU model, or other detailed hardware specifications.
Software Dependencies No The paper mentions
Experiment Setup Yes The paper discusses various aspects of the experimental setup, including hyperparameter effects (Table 1, Table 4), learning rate adjustments ("t SNE s learning rate stays constant over training while UMAP s linearly decreases"), and default settings for GDR ("GDR therefore defaults to t SNE s asymmetric attraction and a and b scalars along with UMAP s distance-metric, initialization, nearest neighbors, and pij symmetrization").