Anti-bias ML for computer vision
Current facial recognition and pose detection computer vision solutions have bias against certain people groups. This affects people worldwide, as these technologies are used in more settings, from law enforcement to health care, and affect people all over the world. As advanced technologies become more affordable in cheap smart phones with Internet access, the number of people who are affected by this keeps increasing. Current facial recognition and pose detection computer vision solutions have bias against certain people groups. This affects people worldwide, as these technologies are used in more settings, from law enforcement to health care, and affect people all over the world. As advanced technologies become more affordable in cheap smart phones with Internet access, the number of people who are affected by this keeps increasing.
Current facial recognition and pose detection computer vision solutions have bias against certain people groups. This affects people worldwide, as these technologies are used in more settings, from law enforcement to health care, and affect people all over the world. As advanced technologies become more affordable in cheap smart phones with Internet access, the number of people who are affected by this keeps increasing. Current facial recognition and pose detection computer vision solutions have bias against certain people groups. This affects people worldwide, as these technologies are used in more settings, from law enforcement to health care, and affect people all over the world. As advanced technologies become more affordable in cheap smart phones with Internet access, the number of people who are affected by this keeps increasing.
My solution would initially use nonlinear regression analysis and deep learning to detect if there is any significant bias in current computer vision solutions for facial recognition and pose detection. Secondly, I would augment and develop alternate solutions for facial recognition and pose detection by analyzing data sets with weighted/geometric averages to account for small samples from groups that the technologies show bias against as well as by using worst-case analysis instead of average-case analysis to improve the fairness and performance of these computer vision solutions for more people groups.
My solution would initially use nonlinear regression analysis and deep learning to detect if there is any significant bias in current computer vision solutions for facial recognition and pose detection. Secondly, I would augment and develop alternate solutions for facial recognition and pose detection by analyzing data sets with weighted/geometric averages to account for small samples from groups that the technologies show bias against as well as by using worst-case analysis instead of average-case analysis to improve the fairness and performance of these computer vision solutions for more people groups.
- Actively minimize human and algorithmic biases, particularly in healthcare, education, and workplace settings.
My solution would initially use nonlinear regression analysis and deep learning to detect if there is any significant bias in current computer vision solutions for facial recognition and pose detection. Secondly, I would augment and develop alternate solutions for facial recognition and pose detection by analyzing data sets with weighted/geometric averages to account for small samples from groups that the technologies show bias against as well as by using worst-case analysis instead of average-case analysis to improve the fairness and performance of these computer vision solutions for more people groups.
- Prototype: A venture or organization building and testing its product, service, or business model.
Prototype. I am testing the prototype that I have developed.
- A new technology
Yes, it uses transfer learning and reinforcement learning to concurrently detect and mitigate bias in machine learning -based solutions for computer vision tasks of facial recognition and pose detection.
- Artificial Intelligence / Machine Learning
- Software and Mobile Applications
- Women & Girls
- Pregnant Women
- LGBTQ+
- Infants
- Children & Adolescents
- Elderly
- Rural
- Peri-Urban
- Urban
- Poor
- Low-Income
- Middle-Income
- Refugees & Internally Displaced Persons
- Minorities & Previously Excluded Populations
- Persons with Disabilities
- 3. Good Health and Well-being
- 5. Gender Equality
- 10. Reduced Inequality
- 16. Peace and Justice Strong Institutions
By measuring the differences in scores for each metric in facial recognition and pose detection, and report their relative difference.
By report the difference for the average and worst-case score for each set of people group (or combination of identities), we can track if these scores are comparable/fair.
- Not registered as any organization
1
I am a Ph.D. student at Texas A&M University's Department of Electrical and Computer Engineering, working on domain-specific computing for computer vision, via hardware/software co-design.
I plan to recruit people from Latinx and Black American student organizations that I am part of.
I aim to emphasize discussing inclusive diversity, equity, and accessibility while recruiting people.
- Organizations (B2B)
To get access to advice from experts and partners of MIT Solve to address U.N. sustainable development goals in computer vision and machine learning solutions that I develop.
- Human Capital (e.g. sourcing talent, board development, etc.)
- Financial (e.g. improving accounting practices, pitching to investors)
- Monitoring & Evaluation (e.g. collecting/using data, measuring impact)
- Technology (e.g. software or hardware, web development/design, data analysis, etc.)
I need help to acquire a large data sets of videos and photos of people from different marginalized people groups to adequately validate and test my machine learning models for facial recognition and pose detection.
Also, I need help to refine my solutions with automated machine learning to efficient perform design space exploration of computer vision solutions.
MIT Faculty from CSAIL who work on bias, fairness, accountability, and transparency in machine learning. They can provide technical advice when I get stuck on local minimals in solution refinement, need to explore other techniques to address the problem.
- Yes, I wish to apply for this prize
The solution that we are prototyping and testing is an anti-racist technology for computer vision, with applications in health care; hence, it address health care inequity as a consequence.
- No, I do not wish to be considered for this prize, even if the prize funder is specifically interested in my solution
- Yes, I wish to apply for this prize
It addresses racial bias in computer vision solutions for facial recognition and pose detection.
- Yes, I wish to apply for this prize
The technology that we are developing is anti-racist, and can mitigate racist policies in law enforcement, social services, and health care. Hence, by mitigating inequities in these areas, the technology contributes to smart, safe, and sustainable communities by improving fairness in arrests and sentencing, medical diagnosis and treatment, and allowing more healthy people to live in communion with family and friends for longer lifespans.
- Yes, I wish to apply for this prize
Our anti-racist technology in computer vision improves inclusive diversity and economic opportunity in communities in the U.S. and around the world, since a more fair law enforcement process and more equitable health care allows people to access more educational, employment, and investment opportunities.
- Yes, I wish to apply for this prize
The technology detects and mitigates bias for multiple people groups, including disabled women of color. Hence, it tests for bias against different sets of intersectional identities, such as the aforementioned category.
Hence, this technology is feminist.
- Yes, I wish to apply for this prize
The anti-racist, feminist, and anti-ableist computer vision technology realizes applied machine learning for all in applications such as law enforcement and health care.