Difference between revisions of "Fair representation learning of electronic health records"
| Line 3: | Line 3: | ||
|Keywords=fair machine learning, bias, fairness | |Keywords=fair machine learning, bias, fairness | ||
|TimeFrame=Fall 2022 | |TimeFrame=Fall 2022 | ||
| − | |References=Reddy, C., Sharma, D., Mehri, S., Romero-Soriano, A., Shabanian, S. and Honari, S., 2021, June. Benchmarking bias mitigation algorithms in representation learning through fairness metrics. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1). | + | |References=Dullerud, N., Roth, K., Hamidieh, K., Papernot, N. and Ghassemi, M., 2022. Is fairness only metric deep? evaluating and addressing subgroup gaps in deep metric learning. arXiv preprint arXiv:2203.12748. |
| + | |||
| + | Reddy, C., Sharma, D., Mehri, S., Romero-Soriano, A., Shabanian, S. and Honari, S., 2021, June. Benchmarking bias mitigation algorithms in representation learning through fairness metrics. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1). | ||
Yuan, Y., Xun, G., Suo, Q., Jia, K. and Zhang, A., 2019. Wave2vec: Deep representation learning for clinical temporal data. Neurocomputing, 324, pp.31-42. | Yuan, Y., Xun, G., Suo, Q., Jia, K. and Zhang, A., 2019. Wave2vec: Deep representation learning for clinical temporal data. Neurocomputing, 324, pp.31-42. | ||
| Line 11: | Line 13: | ||
The ultimate goal of this project is to analyze the effect of different EHR representation learning methods on fairness and look for representations that are agnostic to the patients’ sensitive attributes or have low subgroup gaps in downstream tasks. | The ultimate goal of this project is to analyze the effect of different EHR representation learning methods on fairness and look for representations that are agnostic to the patients’ sensitive attributes or have low subgroup gaps in downstream tasks. | ||
| + | |||
| + | Potential datasets: | ||
| + | https://mimic.mit.edu/docs/about/ | ||
| + | https://docs.nightingalescience.org/ | ||
Revision as of 10:26, 3 October 2022
| Title | Fair representation learning of electronic health records |
|---|---|
| Summary | Fair representation learning of electronic health records |
| Keywords | fair machine learning, bias, fairness |
| TimeFrame | Fall 2022 |
| References | Dullerud, N., Roth, K., Hamidieh, K., Papernot, N. and Ghassemi, M., 2022. Is fairness only metric deep? evaluating and addressing subgroup gaps in deep metric learning. arXiv preprint arXiv:2203.12748.
Reddy, C., Sharma, D., Mehri, S., Romero-Soriano, A., Shabanian, S. and Honari, S., 2021, June. Benchmarking bias mitigation algorithms in representation learning through fairness metrics. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1). Yuan, Y., Xun, G., Suo, Q., Jia, K. and Zhang, A., 2019. Wave2vec: Deep representation learning for clinical temporal data. Neurocomputing, 324, pp.31-42. |
| Prerequisites | |
| Author | |
| Supervisor | Ali Amirahmadi, Ece Calikus, Kobra Etminani |
| Level | |
| Status |
Deep representation learning methods have shown promising performance in different domains, including NLP, image analysis, and healthcare modeling. In healthcare, researchers focus on mitigating data sparsity and high dimensionality and modeling the complex short and long-term dependencies in electric health records (EHR by different representation learning methods. These models are used as feature extractors for few-shot learning and various other downstream tasks. Ensuring fairness in machine learning is extremely important to achieve health equity across different groups in society. Different approaches and frameworks address bias and fairness issues, such as anti-classification, parity, and calibration.
The ultimate goal of this project is to analyze the effect of different EHR representation learning methods on fairness and look for representations that are agnostic to the patients’ sensitive attributes or have low subgroup gaps in downstream tasks.
Potential datasets: https://mimic.mit.edu/docs/about/ https://docs.nightingalescience.org/
