Tiffany Ding

Statistics PhD student @UCBerkeley

[Tiffany Ding @tifding]

Graduated! Very grateful for my advisors, Mike Jordan and Ryan Tibshirani, who set the Pareto frontier for brilliance and kindness 🧠🩷

Next stop: MIT for a postdoc with @stats_stephen!

[Tiffany Ding @tifding]

Love you, mom 😭

[Tiffany Ding @tifding]

Come say hi at my @iclr_conf poster (Friday 10:30am, Pavilion 3 #316) to chat about prediction sets or why @PlantNetProject is my favorite plant identification app!

[Tiffany Ding @tifding]

New paper: Conformal prediction for long-tailed classification🐒
https://t.co/J1PJARDU3M

🧑‍🌾 (plant enthusiast): Help me identify plants!
🤖 (existing conformal algs): Do you want sets that never include rare plants or sets that contain 100s of labels?
🧑‍🌾: Uhh… neither? A🧵(1/n)

[Tiffany Ding @tifding]

My collaborator Joseph Salmon made a beautiful blog post about our paper, featuring an *interactive* recreation of a figure from Sadinle, Lei & Wasserman's iconic 2019 conformal prediction paper. Go check it out!

Blog post: https://t.co/IBp17upNQa

[Tiffany Ding @tifding]

New paper: Conformal prediction for long-tailed classification🐒
https://t.co/J1PJARDU3M

🧑‍🌾 (plant enthusiast): Help me identify plants!
🤖 (existing conformal algs): Do you want sets that never include rare plants or sets that contain 100s of labels?
🧑‍🌾: Uhh… neither? A🧵(1/n)

[Tiffany Ding @tifding]

@kayo_yin So fun biking with you! More rides soon… Just found out that Wildcat Canyon Rd is being closed to cyclists so it can be reopened for cars — tragic. Have to take advantage of good biking while we can :)

[Tiffany Ding @tifding]

This is joint work with Jean-Baptiste Fermanian & Joseph Salmon and inspired by discussions with the rest of the @PlantNetProject team (p.s. Check out their awesome plant identification app if you haven’t yet!🌱) (n/n)

[Tiffany Ding @tifding]

New paper: Conformal prediction for long-tailed classification🐒
https://t.co/J1PJARDU3M

🧑‍🌾 (plant enthusiast): Help me identify plants!
🤖 (existing conformal algs): Do you want sets that never include rare plants or sets that contain 100s of labels?
🧑‍🌾: Uhh… neither? A🧵(1/n)

[Tiffany Ding @tifding]

All of these tools lead to strong empirical performance on PlantNet and iNaturalist. Check out the looooong tail; this is a tough setting! Some classes are left with 0 holdout examples for calibration (4/n)

[Jivat Kaur @jivatneet]

📯 New work! Conformal Prediction Sets with Improved Conditional Coverage using Trust Scores.

https://t.co/fWBfRIObOf

How useful are prediction sets that achieve 90% marginal coverage by failing on 10% cases that are challenging? Not very useful for clinicians who require uncertainty estimates for critical cases in diagnosis!

Can we improve coverage where it matters most? 🧵 (1/n)

[Anastasios Nikolas Angelopoulos @ml_angelopoulos]

📣Announcing the 2024 NeurIPS Workshop on Statistical Frontiers in LLMs and Foundation Models 📣

Submissions open now, deadline September 15th

https://t.co/mJeNkWQv2U

If your work intersects with statistics and black-box models, please submit! This includes:
✅ Bias
✅ Benchmarking
✅ Automatic evaluation
✅ Watermarking
✅ Conformal prediction and black-box uncertainty quantification
✅ Privacy and data rights
✅ Auditing

We’re looking forward to seeing your submissions!
Co-organizers: @stats_stephen @alexdamour @tatsu_hashimoto @JessicaHullman @FannyYangETH !
Program Committee: Chance Johnstone @DrewPrinster @huiwensun_ Eleni Straitouri @MargauxZaffran
Provisional speakers include: @MihaelaVDS @bschoelkopf Virginia Smith and @weijie444 , up to scheduling considerations :)

[Tiffany Ding @tifding]

@JessicaHullman Thanks for the post! I’m curious what you think the conformal community (i.e., ppl working on methods + theory) could do to complement the great work being done by your group and others in HCI to understand how humans decision makers actually use conformal prediction sets

[Tiffany Ding @tifding]

@drKuangzheng @UCBStatistics @SFBART @aashen12 @drKuangzheng We used Adobe Illustrator!

[Tiffany Ding @tifding]

The greatest accomplishment of my statistics career has been winning this year’s @UCBStatistics T-shirt design competition with a @SFBART-inspired shirt designed w/ @aashen12!

{stats nerds} ∩ {public transit nerds} ≠ ∅ 📉🚅

[Tiffany Ding @tifding]

@jjcherian Thanks John! The shirts probably won’t be ready in time for this semester’s colloquium but maybe next fall 😄

[Tiffany Ding @tifding]

Sorry in advance if we missed topics! We tried to be as comprehensive as possible given the space constraints

[Tiffany Ding @tifding]

Interested in uncertainty quantification and how to make conformal prediction sets more practically useful? Come to our poster at #NeurIPS23!

📍Poster #1623
🕙 Thursday 10:45-12:45
w/ @ml_angelopoulos, @stats_stephen, Michael I. Jordan & Ryan Tibshirani

https://t.co/t8QZT827j0

[Tiffany Ding @tifding]

This is joint work with @ml_angelopoulos, @stats_stephen, Michael I. Jordan, and Ryan J. Tibshirani.

Please reach out if you’re interested in chatting about conformal prediction or statistics in general! (4/4)

[Tiffany Ding @tifding]

📢New paper! Class-conditional conformal prediction with many classes

https://t.co/t8QZT827j0

👩‍⚕️ (doctor): “Give me uncertainty on the patient diagnosis.”
🤖(conformal): “The 95% prediction set is {normal}.”
👨🏻(patient): ☠️

How do you avoid this negative outcome? A 🧵(1/4)

[Tiffany Ding @tifding]

A naive strategy is to split the data classwise and run conformal once per class.

But with many classes/limited data, this gives bad results (big sets, etc.)

In clustered conformal prediction, we cluster classes that have similar score distributions and pool their data! (3/4)