Your Classifier can Secretly Suffice Multi-Source Domain Adaptation

Authors: Naveen Venkat, Jogendra Nath Kundu, Durgesh Singh, Ambareesh Revanur, Venkatesh Babu R

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive evaluation of our approach over five benchmark datasets, with two popular CNN backbone models (Res Net-50, Res Net-101 [16]) and derive insights from the empirical analysis.
Researcher Affiliation Academia Naveen Venkat Jogendra Nath Kundu Durgesh Kumar Singh Ambareesh Revanur R. Venkatesh Babu Video Analytics Lab, Indian Institute of Science, Bangalore
Pseudocode Yes Algorithm 1 SImp Al Self-supervised Implicit Alignment
Open Source Code Yes See code implementation1 for architecture, hyperparameter values, and instructions to reproduce the results. 1http://val.cds.iisc.ac.in/simpal
Open Datasets Yes Office-31 [47] dataset... Image CLEF2 dataset... Image Net (I) [46], Caltech-256 (C) [14], Pascal-VOC 2012 (P) [9]... Office-Caltech [12] dataset... Office-Home [59]... Domain Net [43] dataset...
Dataset Splits Yes For Image CLEF and Office-based datasets, we follow the evaluation protocol in MFSAN [68], while for Domain Net, we follow the protocol used in M3SDA [43].
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models or memory specifications used for running the experiments.
Software Dependencies No The paper states 'We implement our approach in Py Torch [41]' but does not specify the version number of PyTorch or any other software dependencies.
Experiment Setup Yes We use the Adam [19] optimizer, with learning rate 10 5 and weight decay 5 10 4 for stochastic optimization. The losses in Eq. 3 and Eq. 7 are alternatively optimized and the target agreement rate (Eq. 5) is periodically monitored for convergence. We set ne = 15 epochs as the update rate for the target pseudo-labels (line 12 in Algo. 1).