process, causing me to lose time, and all my important data and work. After the last repair failed just one day later, I have no choice but to buy a new laptop to continue my work. As a grad student, this has been a huge setback. I hope you will consider appropriate compensation.

Finally, we experimentally validate our theoretical results. Work done during an internship at @GoogleDeepMind, along with Yukti Makhija, Rishi Saket, and Aravindan Raghuveer.

with respect to natural objectives such as k-means. We also show that our bagging mechanisms can be made label-differentially private, incurring an additional utility error. We then generalize our results to the setting of Generalized Linear Models (GLMs).

such that the utility for downstream tasks like linear regression is maximized. We theoretically provide utility guarantees, and show that in each case, the optimal bagging strategy (approximately) reduces to finding an optimal clustering of the feature vectors or the labels,

In the case of MIR, the bag-label is the label of an undisclosed instance from the bag, while in LLP, the bag-label is the mean of the bag's labels. In this paper, we study for various loss functions in MIR and LLP, what is the optimal way to partition the dataset into bags,

Multiple Instance Regression (MIR) and Learning from Label Proportions (LLP) are useful learning frameworks, where the training data is partitioned into disjoint sets or bags, and only an aggregate label, i.e., bag-label for each bag is available to the learner.

Will present "Aggregating Data for Optimal Learning" at #UAI2025 in Brazil, on 22nd July (today!). https://t.co/Y0md3SUtae ⏲️Oral presentation: 11:30am ⏲️Poster: 4pm Please have a look, and do stop by if it sounds interesting to you! RT's appreciated😊Summary to follow

Work done along with Amit Deshpande, Rajmohan Rajaraman, and Ravi Sundaram https://t.co/LhPEbnklHG

We also demonstrate that randomization leads to better accuracy and efficiency. However, we show that the randomized Fair BOC is nearly-deterministic, giving randomized predictions on at most one data point, hence availing benefits of randomness, while using very little of it.

Hence, we consider the randomized Fair BOC, and our central result is that its accuracy is robust to malicious noise in the data distribution. Our robustness result applies to various fairness constraints---Demographic Parity, Equal Opportunity, Predictive Equality.

Kearns & Li (1988) implies that the accuracy of the deterministic BOC without any fairness constraints is robust (Lipschitz) to malicious noise in the data distribution. We demonstrate that their robustness guarantee breaks down when we add fairness constraints.

Previous work in fair machine learning has characterised the Fair Bayes Optimal Classifier (BOC) on a given distribution for both deterministic and randomized classifiers. We study the robustness of the Fair BOC to adversarial noise in the data distribution.

Presenting "Optimal Fair Learning Robust to Adversarial Distribution Shift" at #ICML2025 ( https://t.co/LhPEbnjNS8) 📍East Exhibition Hall A-B #E-1001 ⏲️16th July, 4:30-7PM Please have a look, and do stop by if it sounds interesting to you! RT's appreciated😊Summary to follow

Previous work in fair machine learning has characterised the Fair Bayes Optimal Classifier (BOC) on a given distribution for both deterministic and randomized classifiers. We study the robustness of the Fair BOC to adversarial noise in the data distribution.

I’m happy to announce that our paper, “Market or Markets? Investigating Google Search's Market Shares Under Vertical Segmentation,” won Best Paper Honorable Mention at @icwsm with @jeffreylgleason , @whocheers @bowlinearl , @RERobertson and @nikenberger https://t.co/Yt02LeQFzd

Work done during an internship at @MSFTResearch, along with my internship mentor Amit Deshpande

We construct fair representations on a given data distribution that have provably optimal accuracy and suffer no accuracy loss compared to the optimal fair classifiers on the original data distribution! Fairness notions: Demographic Parity, Equal Opportunity, Predictive Equality

Previous techniques for fair representations on a distribution are mainly heuristic. The classifiers on these representations either come with no or weak accuracy guarantees when compared to the optimal fair classifier on the original distribution. We improve on these works!

Previous work on fair classification has characterized the optimal fair classifiers on a given data distribution. We refine these characterizations to demonstrate when the optimal *randomized* fair classifiers can surpass their deterministic counterparts in accuracy!

Found out that my @FAccT 2022 paper on Fair Representation Learning hasn't been accessible on the ACM library webpage for a long while (paywall😥) Just posted it to arXiv, better late than never😅 Please have a look! RT's appreciated😊Summary to follow https://t.co/crh7yJWAMS