Regression under Human Assistance

Authors: Abir De, Paramita Koley, Niloy Ganguly, Manuel Gomez-Rodriguez2611-2620

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic and real-world data from two important applications medical diagnosis and content moderation demonstrate that the greedy algorithm beats several competitive baselines. Finally, we experiment with synthetic and real-world data from two important applications medical diagnosis and content moderation. Our results show that the greedy algorithm beats several competitive algorithms, including the iterative algorithm for maximization of a difference of submodular functions mentioned above, and is able to identify and outsource to humans those samples where their expertise is required.
Researcher Affiliation Academia Abir De MPI-SWS ade@mpi-sws.org Paramita Koley IIT Kharagpur paramita.koley@iitkgp.ac.in Niloy Ganguly IIT Kharagpur niloy@cse.iitkgp.ac.in Manuel Gomez-Rodriguez MPI-SWS manuelgr@mpi-sws.org
Pseudocode Yes Algorithm 1 Greedy algorithm Input: Ground set V, set of training samples {(xi, yi)}i V, parameters n and λ. Output: Set of items S 1: S 2: while |S| < n do 3: % Find best sample 4: k argmaxk V\S log ℓ(S k) + log ℓ(S) 5: % Sample is outsourced to humans 6: S S {k } 7: end while 8: return S
Open Source Code Yes To facilitate research in this area, we are releasing an open source implementation of our method1. 1https://github.com/Networks-Learning/regression-underassistance
Open Datasets Yes We experiment with four real-world datasets from two important applications, medical diagnosis and content moderation, which are publicly available (Davidson et al. ; Decenci ere et al. 2014; Hoover, Kouznetsova, and Goldbaum 2000).
Dataset Splits No Finally, in each experiment, we use 80% samples for training and 20% samples for testing. The paper does not explicitly mention a separate validation split.
Hardware Specification No The paper does not provide specific hardware details (such as GPU/CPU models or memory amounts) used for running its experiments.
Software Dependencies No The paper mentions using fasttext and Resnet for feature extraction but does not provide specific version numbers for any software components or dependencies.
Experiment Setup Yes Experimental setup. For each sample (x, y), we first generate each dimension of the feature vector x Rd uniformly at random, i.e., xi U( 1, 1) and then sample the response variable y from either (i) a Gaussian distribution N(1 x/d, σ2 1) or (ii) a logistic distribution 1/(1 + exp( 1 x/d)). Moreover, we sample the associated human error from a Gaussian distribution, i.e., c(x, y) N(0, σ2 2). In each experiment, we use |V| = 500 training samples and we compare the performance of the greedy algorithm with three competitive baselines:. In panel (a), we set λ = 5 10 3 and, in panel (b), we set λ = 10 3.