Ryan
@ryanboustany.bsky.social
PhD candidate in ML - @thalesgroup & @TSEinfo & @ANITI_Toulouse.
💡 How much more training is needed to move from a biased to a fair predictor ?
Empirically: up to 400% extra training.
Joint work with François Bachoc, Jérôme Bolte and Jean-Michel Loubes.
(3/3)
Empirically: up to 400% extra training.
Joint work with François Bachoc, Jérôme Bolte and Jean-Michel Loubes.
(3/3)
May 23, 2025 at 4:48 PM
💡 How much more training is needed to move from a biased to a fair predictor ?
Empirically: up to 400% extra training.
Joint work with François Bachoc, Jérôme Bolte and Jean-Michel Loubes.
(3/3)
Empirically: up to 400% extra training.
Joint work with François Bachoc, Jérôme Bolte and Jean-Michel Loubes.
(3/3)
- The majority loss dominates the landscape: minimizing the loss is almost always equivalent to minimizing the majority part.
- We introduce a stereotype gap: the distance between fair and majority-optimal critical points.
- Debiasing requires extra training.
(2/3)
- We introduce a stereotype gap: the distance between fair and majority-optimal critical points.
- Debiasing requires extra training.
(2/3)
May 23, 2025 at 4:48 PM
- The majority loss dominates the landscape: minimizing the loss is almost always equivalent to minimizing the majority part.
- We introduce a stereotype gap: the distance between fair and majority-optimal critical points.
- Debiasing requires extra training.
(2/3)
- We introduce a stereotype gap: the distance between fair and majority-optimal critical points.
- Debiasing requires extra training.
(2/3)