Towards Socially Responsible AI: Cognitive Bias-Aware Multi-Objective Learning
Authors: Procheta Sen, Debasis Ganguly2685-2692
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments conducted on an emotion prediction task with balanced class priors shows that a set of baseline bias-agnostic models exhibit cognitive biases with respect to gender, such as women are prone to be afraid whereas men are more prone to be angry. In contrast, our proposed bias-aware multi-objective learning methodology is shown to reduce such biases in the predictid emotions. |
| Researcher Affiliation | Collaboration | Procheta Sen,1 Debasis Ganguly2 ADAPT Centre, Dublin City University, Ireland,1 IBM Research, Dublin, Ireland2 |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described in this paper. |
| Open Datasets | Yes | The dataset that we use in particular for our experiments is the Equity Evaluation Corpus (EEC), compiled by the work in (Kiritchenko and Mohammad 2018). |
| Dataset Splits | Yes | In all our experiments, we used a train-test split of 80:20. |
| Hardware Specification | No | The paper does not explicitly describe the hardware used to run its experiments. |
| Software Dependencies | No | The paper mentions using skipgram and logistic regression but does not provide specific software library names with version numbers. |
| Experiment Setup | Yes | We set d = 300 for all our experiments. All approaches, except Bias-Agnstc-DWE , use 300 dimensional pre-trained skipgram vectors trained on Google News corpus. The vector representation of each sentence is the sum of embedded representations of constituent words of the sentence. We tuned the dimension of the shared layer in a range of 10 to 250 in steps of 10, and report the results only with the optimal value of p = 200. In all our experiments, we used a train-test split of 80:20. Specifically, for our experiments, we use square loss to back-propagate errors and compute gradients. |