Berk Ustun @berkustun Twitter profile

Pinned Tweet

Berk Ustun

6 years

Denied a loan by an ML model? You should be able to change something to get approved! In a new paper w @AlexanderSpangh & @yxxxliu , we call this concept "recourse" & we develop tools to measure it for linear classifiers. PDF CODE

4

46

194

Last Seen Profiles

@S4MUNIGIRI

@MattLM2596

@merryfight

@3BallJay

@cronicadeapolo

@CoachCortez58

@ForeyChery

@LillyES

@coronaurajoho

@frankcdilorenzo

@atari_f3

@minsikko

@ColbergLatigo

@Elizabe48431564

@baby_pepes

@nathan_h_b

@Rayxzn_

@Appakhal

@sun9394917

@dunlap_kade

@MomoGlobalUnion

@GonzaMaril32730

@SandersMic13788

@UpNorthCBC

@carlsondavis3

@A10Stats

@J0HNTENIST

@trhk_ran

@boid86

@LaurenWebb48009

@EPAFootball

@zacharyebell

@YoThatsNiffty

@Opacusz

@mpigaliza

@melowmanie

Berk Ustun

@berkustun

3 years

I am thrilled to announce that – after spending the past 16 years in the US, earning a BS+MS+PhD in STEM, and co-founding a startup that employs many Americans – I have finally been granted the right to permanently reside in the United States... as a result of marriage 🎉🗽🇺🇸

177

2K

7K

Berk Ustun

@berkustun

6 months

📢 Please RT 📢 I am recruiting PhD students to join my group at UCSD! We develop methods for responsible machine learning - with a focus on fairness, interpretability, robustness, and safety. Check out for more information.

5

130

353

Berk Ustun

@berkustun

5 years

Early fears of AI automation circa 1933. h/t Creditworthy: A History of Consumer Surveillance by @joshlauer

4

148

321

Berk Ustun

@berkustun

4 years

Personal Update: I'll be starting as an Assistant Professor at @HDSIUCSD in Fall 2021! Until then, I will be @GoogleAI where I'll be working on fairness and interpretability for machine learning in healthcare!

22

6

223

Berk Ustun

@berkustun

6 years

“Please don’t do explainable ML.” Incredible 10 min talk by Cynthia Rudin. Link: (Fast forward to 20 min)

8

35

153

Berk Ustun

@berkustun

6 years

Key proposal by @bhecht in Nature this week: “The CS community should change its peer-review process to ensure that researchers disclose any possible negative societal consequences of their work in papers, or risk rejection.”

5

50

86

Berk Ustun

@berkustun

6 months

📢 Please RT!📢 We're hiring postdoctoral researchers to work on responsible machine learning at UCSD! Topics include fairness, explainability, robustness, and safety. For more, see

2

36

80

Berk Ustun

@berkustun

6 years

@randal_olson If you’re worried about breaking code by switching to Python 3, check out this short guide on GitHub. It has a nice list of the main differences between Python 2/3, which makes switching *much* easier.

GitHub - arogozhnikov/python3_with_pleasure: A short guide on features of Python 3 with examples

A short guide on features of Python 3 with examples - arogozhnikov/python3_with_pleasure

github.com

0

38

72

Berk Ustun

@berkustun

4 years

I really loved this article and this take. The more I work in this field, the more I think that the real issue about human-facing ML isn't "bias" but "power" 1/2

Kristian Lum

@KLdivergence

4 years

“Biased algorithms are easier to fix than biased people”. But, in some cases, I don't see these as entirely different problems. Sometimes fixing a biased algorithm first requires fixing biased people.

3

64

219

1

15

72

Berk Ustun

@berkustun

6 years

Cool new paper to check out if you’re working on interpretable ML: “A review of possible effects of cognitive biases on interpretation of rule-based machine learning models.”

A review of possible effects of cognitive biases on the...

While the interpretability of machine learning models is often equated with their mere syntactic comprehensibility, we think that interpretability goes beyond that, and that human interpretability...

arxiv.org

1

16

51

Berk Ustun

@berkustun

6 years

Excited to announce FAT/ML 2018 in on! This year we’re co-located with @icmlconf in Stockholm 🇸🇪. Call for Papers (due 5/1): @mrtz @KLdivergence @ruggieris @timnitGebru

0

21

46

Berk Ustun

@berkustun

2 years

What are your best tips for running a graduate research seminar online? Think ~20 PhD students reading and discussing ML research papers. Looking for any and all recommendations to make it more engaging and rewarding - course policies, tools, etc.

12

3

47

Berk Ustun

@berkustun

10 months

Machine learning models often use group attributes like sex, age, and race for personalization In our latest work, we show that personalization can lead to worsenalization for certain groups Joint @VMSuriyakumar @MarzyehGhassemi 🧵👇

Vinith M Suriyakumar

@VMSuriyakumar

10 months

Personalized models often use group attributes like sex/age/race. In our latest w @berkustun @MarzyehGhassemi , we show how personalization can lead to worsenalization by reducing performance for some groups PDF #ICML23 Oral

3

25

78

1

10

47

Berk Ustun

@berkustun

4 years

@seanjtaylor Worth checking out "Why Model?" by Joshua Epstein. The optimizer approach makes sense of the ultimate purpose of the model is something other than prediction.

1

14

43

Berk Ustun

@berkustun

6 months

Excited for #NeurIPS2023 and #ML4H2023 ! I'll be in town all of next week. Shoot me an e-mail if you want to catch up or hang out. Also recruiting postdocs and PhD students this year! If you're looking to work on fairness, explainability, robustness, safety, let's chat 👇

Berk Ustun

@berkustun

6 months

📢 Please RT!📢 We're hiring postdoctoral researchers to work on responsible machine learning at UCSD! Topics include fairness, explainability, robustness, and safety. For more, see

2

36

80

0

2

37

Berk Ustun

@berkustun

6 years

Algorithms can outperform humans, but algorithmic decision-making can be unjust due to data collection and seemingly harmless modeling decisions. When we automate consequential decisions, ML practitioners become policy-makers. The brouhaha is about how to do it right.

Pedro Domingos

@pmddomingos

6 years

Algorithms are much less biased than humans, but you'd never know from the current brouhaha about them:

26

234

518

2

8

33

Berk Ustun

@berkustun

5 months

Personalized models should let users consent to the use of their personal information! In our latest, we describe how to build models that let users consent to the use of group attributes like sex, age, race, HIV status Spotlight Poster @NeurIPSConf : Tues 10:45-12:45 PM Link:

Hailey Joren

@HaileyJoren

5 months

Models are trained on costly data and require this data at prediction time. We should be able to opt-out and understand the gains of opting in! In our latest w @nagpalchirag @kat_heller @berkustun we introduce models that give users this informed consent #NeurIPS2023 Spotlight

1

34

0

5

30

Berk Ustun

@berkustun

6 years

parity ≠ fairness @katecrawford @hannawallach at #FATML

2

6

31

Berk Ustun

@berkustun

1 year

15K abstracts @NeurIPSConf means - 10K submissions (30% drop) - 30K reviews (3 per sub) - 5-10K reviewers (3 to 6 subs per reviewer) Do we really have this many reviewers in ML?

Kate Saenko

@kate_saenko_

1 year

15,225 abstracts submitted to @NeurIPSConf 2023 😲

4

10

140

5

1

28

Berk Ustun

@berkustun

3 years

@tomgoldsteincs gpu_system_update (for when you need all of them)

0

1

28

Berk Ustun

@berkustun

4 years

Ask yourself - who is making decisions in algorithm design and model selection? Did I have a say? Were there other reasonable alternative algorithms and models that would have benefitted me? 2/2

2

4

28

Berk Ustun

@berkustun

10 months

Pretend you are a gentle Professor Emeritus of Computer Science. Write a short, edifying, and kind rebuttal to this sloppy review from an overly confident PhD student: [REVIEW] Here are the key points to include: [YOUR ANGRY REBUTTAL]

0

1

25

Berk Ustun

@berkustun

1 year

Datasets often admit multiple "competing" models that perform almost equally well @JamelleWD 's #AAAI23 paper shows that competing models can assign wildly different risk predictions We develop methods to fit competing models, and to measure the sensitivity of risk estimates 👇

Jamelle Watson-Daniels

@JamelleWD

1 year

Just arrived in DC for #AAAI23 excited to present on Predictive Multiplicity in Probabilistic Classification (work with @berkustun and David Parkes) Oral presentation: Feb 11 at 9:30am ET (ML Bias and Fairness session 1) Poster: Saturday Feb 11 6:15pm ET

3

11

54

0

1

24

Berk Ustun

@berkustun

4 years

As good a time as any to say that I am looking for PhD students and postdocs to join my group @HDSIUCSD . If you're interested in fairness and interpretability in machine learning, please apply! More information here:

1

2

23

Berk Ustun

@berkustun

5 years

@zacharylipton Wait until you hear about the Absent Levels problem in the R Random Forests package:

1

2

23

Berk Ustun

@berkustun

1 year

Just got into New Orleans for #NeurIPS2022 . Let's chat! Will be in town til Saturday.

0

1

22

Berk Ustun

@berkustun

10 months

Folks at #ICML2023 who are interested in new open problems about fairness, data rights, and healthcare: @VMSuriyakumar will be presenting our work on "When Personalization Harms Performance" at 4 PM HST in Ballroom C.

Berk Ustun

@berkustun

10 months

Machine learning models often use group attributes like sex, age, and race for personalization In our latest work, we show that personalization can lead to worsenalization for certain groups Joint @VMSuriyakumar @MarzyehGhassemi 🧵👇

1

10

47

0

1

21

Berk Ustun

@berkustun

5 years

Excited to present our work on actionable recourse at the #NeurIPS AI Ethics workshop tomorrow. Stop by and say hi if you’re around! ⚡️ talk and poster session from 3- 4:30p @ Room 516 AB.

Berk Ustun

@berkustun

6 years

Denied a loan by an ML model? You should be able to change something to get approved! In a new paper w @AlexanderSpangh & @yxxxliu , we call this concept "recourse" & we develop tools to measure it for linear classifiers. PDF CODE

4

46

194

1

3

21

Berk Ustun

@berkustun

5 years

Question for ML folks: Say two classifiers have equal test error (on average), but they weigh their inputs differently and output conflicting predictions on specific points. How do you choose which model to deploy?

7

0

19

Berk Ustun

@berkustun

6 years

@joaoaveiga @pmddomingos I'd also add that they're as biased as the data that they learn from.

1

19

Berk Ustun

@berkustun

4 years

We should measure and report predictive multiplicity like we measure and report test error. Thread 👇

Charlie Marx

@CharlieTMarx

4 years

Denied parole by an ML model? The next best model might have decided otherwise In our #ICML20 paper w @berkustun @FlavioCalmon , we study the ability for an ML problem to admit competing models with conflicting predictions, which we call "predictive multiplicity" THREAD ⬇️

3

23

87

0

3

19

Berk Ustun

@berkustun

3 years

Authors of the triply robust method after seeing this paper:

Fuck Everything, We're Doing Five Blades

Would someone tell me how this happened? We were the fucking vanguard of shaving in this country. The Gillette Mach3 was the razor to own. Then the other guy came out with a three-blade razor. Were...

www.theonion.com

John Mullahy

@JohnMullahy

3 years

🤔

9

4

80

0

19

Berk Ustun

@berkustun

5 years

@hannawallach @ruchowdh @jpatrickhall put together a nice list of interpretability / fairness tools here:

0

3

17

Berk Ustun

@berkustun

5 years

Looking forward to reading this... and to seeing more "Please Stop" papers in ML.

Stat.ML Papers

@StatMLPapers

5 years

Please Stop Permuting Features: An Explanation and Alternatives. (arXiv:1905.03151v1 [])

0

5

29

1

2

18

Berk Ustun

@berkustun

5 years

Checklists are surprisingly effective when they are used to support human decisions (rather than replace it). We should be learning them from data, and making better use of them in domains like medicine.

1

2

17

Berk Ustun

@berkustun

3 years

Some intuition for what we lose with synthetic data generation.... h/t @ustunb

Marzyeh

@MarzyehGhassemi

3 years

My student @itsvictoriaday has a great @FAccTConference paper that shows synthetic differentially private data *doesn't* solve downstream disparities in prediction. We need *real* diverse data!

0

23

1

16

Berk Ustun

@berkustun

6 years

15+ years after the release of the randomForests package... Someone finally discovers a key bug in the way that RFs use categorical input variables.

Journal of Machine Learning Research

@JmlrOrg

6 years

"Random Forests, Decision Trees, and Categorical Predictors: The ``Absent Levels'' Problem", by Timothy C. Au.

0

3

4

0

8

17

Berk Ustun

@berkustun

6 years

I think this might be the first paper on building a fair predictive model for a real-world healthcare application. Awesome!

Stat.ML Papers

@StatMLPapers

6 years

Creating Fair Models of Atherosclerotic Cardiovascular Disease Risk. (arXiv:1809.04663v1 [cs.LG])

0

7

2

5

16

Berk Ustun

@berkustun

9 months

Excited to present our work on worsenalization in machine learning @UCSF_BCHSI next week! Come say hi if you're interested in fairness, data privacy, and clinical prediction models!

UCSF Bakar Computational Health Sciences Institute

@UCSF_BCHSI

9 months

"When personalization harms performance" seminar via @UCSF_Epibiostat Sept. 6, 3pm Berk Ustun @berkustun , Assistant Professor, Dept of Computer Science and Engineering, UC San Diego

0

2

5

1

3

17

Berk Ustun

@berkustun

6 years

@devisridhar “Professor wins award for having so many awards.” Credit to: @Research_Tim

Tim van der Zee

@Research_Tim

6 years

I forced a bot to watch five faculty meetings (approx. 1,000 hours total) and then forced it to write a faculty meeting script of its own. Here is the first page.

120

2K

5K

0

1

16

Berk Ustun

@berkustun

5 years

@kamalikac @rsalakhu The ICML code policy was great this year! As reviewers, we could check the code to answer clarifying questions about methods & datasets As authors, we could afford to leave out tedious details about experiments and point readers to the code Thank you for making it happen.

0

4

16

Berk Ustun

@berkustun

6 years

@PM_1729 I’d say the points are: - interpretable model ≠ blackbox model + post hoc explainer - blackbox model + post hoc explainer = bad There’s an implication that interpretable ML is a solution. Honestly tho, it has limitations. Sometimes it’s better to build no model at all.

1

14

Berk Ustun

@berkustun

4 years

!!!

Page Not Found | Duke Pratt School of Engineering

pratt.duke.edu

0

1

15

Berk Ustun

@berkustun

5 years

Great list of points about the mismatch between research and practice in fair ML by @hannawallach

0

4

15

Berk Ustun

@berkustun

2 years

Startup idea: CS conference but charge $1K/paper to extend the submission deadline by 1 day

1

0

14

Berk Ustun

@berkustun

5 years

@Aaroth Some surveys and position papers to check out: 1. Explanation in AI: Insights from the Social Sciences by @tmiller_unimelb 2. Cognitive Biases on the Interpretation of Rule-Based ML models 1/2

1

0

13

Berk Ustun

@berkustun

5 months

achievement unlocked!

1

0

14

Berk Ustun

@berkustun

5 years

One way to (partially) deal with this issue: require all methods papers @fatconference to include a limitations section.

0

13

Berk Ustun

@berkustun

6 years

@random_walker We should ask for papers to include a “limitations” section imo.

1

3

13

Berk Ustun

@berkustun

6 years

Surprisingly thoughtful white paper from US Congress on the challenges with AI. Nice to see large parts focusing on privacy, fairness, malicious uses, inspectability, and the need for research funding :-)

Jessica Newman

@JessicaH_Newman

6 years

"it’s critical that the federal government address the different challenges posed by AI, including its current and future applications." - new AI White Paper from the U.S. House of Representatives

0

1

6

0

4

13

Berk Ustun

@berkustun

4 years

Just got into Vancouver in time for the #NeurIPS2019 workshops! Let's chat! I'll be at the human-centered ML workshop on Friday, and the Fairness in ML for Healthcare workshop on Saturday.

0

2

13

Berk Ustun

@berkustun

5 years

A thought-provoking piece on what we teach students in CS/ML/OR today Touches on everything from linear programming to neural nets to EC.

1

2

11

Berk Ustun

@berkustun

5 years

Question for the fair ML crowd: Do you include protected attributes (e.g. race, gender) as an input variable for a model that is trained under a fairness constraint (e.g. equalized odds)? Why / why not?

5

1

11

Berk Ustun

@berkustun

1 year

What's the right way to do handle a ICML review stating that your paper doesn't sufficiently expand on a previous version at a workshop? Issues: 1. Workshop paper is non-archival so it shouldn't matter. 2. Reviewer must have googled paper (so no longer blinded).

4

2

11

Berk Ustun

@berkustun

5 months

@MilenaAlmagro @Gaffetheory Great list! Adding a few more tips I wish I had known: - Clif Bars to hold you over - Comfortable shoes so you can walk across campus if needed - List of 10-20 questions for 1:1s when you run out of things to chat about :-) - Remember to ask for breaks between 1:1s

0

11

Berk Ustun

@berkustun

2 years

ML folks - does anyone know the origin story surrounding adversarial examples? Wondering if we can trace the research interest back to a talk/talks or a paper/papers etc.

8

0

11

Berk Ustun

@berkustun

6 years

@mikarv @m_sendhil Cool result! It’s not really relevant to settings where models are learned from data. It’s still a bad idea to fit a complex model that humans cannot understand or validate (e.g. a NN for credit scoring), just because ∃ a complex model that produces fairer allocation decisions.

2

10

Berk Ustun

@berkustun

4 years

Thank you to @HCRCS for hosting me for the last three years. Also, a huge shoutout my brilliant collaborators: @CynthiaRudin , David Parkes, @FlavioCalmon , @roboticwrestler , Ron Kessler, Margo Seltzer, Brandon Westover, @yangl1u , @HW_HaoWang , @AlexanderSpangh , @CharlieTMarx .

0

10

Berk Ustun

@berkustun

4 years

@AlexGDimakis One more failure mode :-) Removing an independent feature could also affect performance by oversimplifying the hypothesis class. Example: X1 = vector of ones used to represent the intercept in a linear classifier. Removing X1 = no more intercept.

0

10

Berk Ustun

@berkustun

4 years

@geomblog @kdphd Computer scientists: "An interesting direction for future work."

0

1

9

Berk Ustun

@berkustun

5 years

Next up at #INFORMS2019 11am - Fairness, Accountability and Transparency in ML Feat: @uhlily @AlexanderSpangh @angelamczhou @CharlieMarx9

𝙷𝚒𝚖𝚊 𝙻𝚊𝚔𝚔𝚊𝚛𝚊𝚓𝚞

@hima_lakkaraju

5 years

If you are going to be at INFORMS Annual Meeting this year, you should not miss our session on Fairness in ML (October 20th, 8am). We have an amazing speaker line up: @5harad , @joftius , @zacharylipton . More info @INFORMS2019 @berkustun #fairml

1

9

23

0

1

9

Berk Ustun

@berkustun

5 months

@geomblog I use IFTTT to post the arXiv RSS feed into a papers and abstracts Slack channel. You can then skim quickly and save the ones that seem relevant into folders for each project. This leads to a pile of PDFs to "read" for each project. And then I go them again when writing

0

1

9

Berk Ustun

@berkustun

6 years

Bad ideas for NY teacher hiring score: - Using “good teacher evals” as an output (bad proxy for student success) - Using “personality traits” as an input (irrelevant features) - Using only data from teachers who were hired (bad sample selection) Also why are we scoring?

0

2

9

Berk Ustun

@berkustun

4 months

ML Conference Cheat Code: - Figure 1 - Definition - Remark - Assumption - Remark - Proposition - Theorem - Corollary, Corollary, Corollary - BIG TABLE OF EXPERIMENTAL RESULTS - Impact Statement

3

0

27

Berk Ustun

@berkustun

6 years

@ruchowdh @KayFButterfield @johnchavens @vdignum @jonniepenn @j2bryson @tabithagold @maria_axente @robmccargow @Debbie_Hellman @fatconference We also associate “explainability” with ideas like “contestability”, “autonomy”, “recourse.” We should speak about the latter ideas separately since explanations in ML guarantee none of these things. Good thread here:

Michael Veale @[email protected]

@mikarv

6 years

Explanations are convenient for large firms in algorithms, AI, ML They don't give users control They don't shake-up power relations They don't shine light on systems as a whole It's irresponsible of researchers to jump on the explanation bandwagon without being critical of them

4

30

93

1

8

Berk Ustun

@berkustun

11 months

On the 🧵 app as @berkmlr ! It's not the same yet but should get there once they release an API (for accounts like @StatMLPapers ), and a critical mass of AI/ML content (so we don't have to see generic posts).

2

1

7

Berk Ustun

@berkustun

6 years

Interested in causal fairness in ML? Here’s a thought-provoking paper by Issa Kohler-Hausmann. “Eddie Murphy and the Dangers of Counterfactual Thinking About Detecting Racial Discrimination”

1

8

Berk Ustun

@berkustun

10 months

@begusgasper Speaking from years of experience, Berk means: - "ew" in France - "idiot" in Britain - "resilient" in Turkey, - and nothing at all in the US :-)

1

0

8

Berk Ustun

@berkustun

6 years

@mikarv Who is funding these clowns? Complaining about the costs associated with “human review” is ridiculous. Companies are already saving $ by automating decisions. The least they could do spend a little to provide consumers with recourse.

1

0

7

Berk Ustun

@berkustun

6 years

@aselbst For datasets: Dataset Nutrition Labels () For models: Automated Bias and Fairness Reports () AI360 Fairness Toolkit: () I’m sure I’m missing lots of others... so looking forward to this thread :-)

AI Fairness 360: An Extensible Toolkit for Detecting,...

Fairness is an increasingly important concern as machine learning models are used to support decision making in high-stakes applications such as mortgage lending, hiring, and prison sentencing....

arxiv.org

1

7

Berk Ustun

@berkustun

5 years

@angelamczhou Maybe for jobs... but for research we should think of it as the kernel trick. It's hard to solve the problem when you stick to one discipline. But then you project it into the higher-dimensional interdisciplinary plane, and boom the problem becomes easy.

1

7

Berk Ustun

@berkustun

5 years

On the top of my summer reading list! Bonus: you can download each chapter as an ePub file for easier reading on a Kindle.

mireillemoret

@mireillemoret

5 years

Dear twitter folk, just now we have launched 'Law for Computer Scientists and Other Folk' with OUP for open review:

9

193

334

0

1

7

Berk Ustun

@berkustun

5 years

Has anyone actually worked on an algorithmic impact assessment? What were you trying to assess? How did it turn out?

Michael Veale @[email protected]

@mikarv

5 years

the danger with many impact assessments is that an impact assessment impact asessment would read 'no impact'

2

6

36

3

0

6

Berk Ustun

@berkustun

5 years

Rebalance your training data! You’d be surprised how often models perform differently across groups (z) due to uneven sample sizes (n) or label imbalances (y) Resampling to equalize n for each (y,z) is a simple fix. It’s perfectly defensible if the training data isn’t iid

Andrew Gelman et al.

@StatModeling

5 years

What is the most important real-world data processing tip you’d like to share with others?

0

2

20

1

0

7

Berk Ustun

@berkustun

4 years

Radient descent

〈 Berger | Dillon 〉

@InertialObservr

4 years

Alter one letter of a concept in your discipline. I'll go first. Dank Matter

645

182

1K

0

7

Berk Ustun

@berkustun

6 years

@mikarv @m_sendhil If we exclusively consider complex models, then fewer stakeholders participate in model selection and we can legitimize decision-making processes that are less just (even they result in fairer allocation)

0

1

7

Berk Ustun

@berkustun

3 years

@yonashav @KLdivergence policymakers: "can we set it to improve performance?"

0

6

Berk Ustun

@berkustun

11 months

What are the best resources to learn about proposed transparency and explainability requirements in the EU AI Act?

1

0

6

Berk Ustun

@berkustun

6 years

@Aaroth @bhecht Hmm you may be right. Part of the issue is that some work has no immediate societal consequence. I do think that papers should include a limitations section tho (which is standard in medicine). studies). It would help peer review / avoid misunderstandings by journalists etc.

1

0

6

Berk Ustun

@berkustun

5 years

@KalaiRamea leaving this here since it always surprises me: “It’s worth recalling that the word ‘meritocracy’ was coined as a satirical slur in a dystopic novel by a sociologist.” h/t @antoniogm

0

2

6

Berk Ustun

@berkustun

5 years

"Citizens may be able to accept algorithmic policy losses if they have more purchase on the rationales behind them and recourse to change them."

Ellen P. Goodman

@ellgood

5 years

Part 2 of my piece on Boston Public Schools’ #algorithm fail and what we can learn about public engagement with the model.

1

3

0

1

6

Berk Ustun

@berkustun

4 years

@JessicaHullman Also shameless plug to some of our work. We developed methods that let domain experts to specify constraints on model form and predictions, and that inform customization by telling them how their constraints affect performance: 2/n

1

6

Berk Ustun

@berkustun

5 years

@tdietterich @roydanroy @mark_riedl @nickfrosst @umangsbhatt @CynthiaRudin Here’s a way to define “interpretable” without using “explanation”: Interpretable = you understand how the model operates. If a model is interpretable: 1. You can print it on a piece of paper 2. You know how its prediction will change if you change any input

1

5

Berk Ustun

@berkustun

3 years

@rajiinio @geomblog @Aaron_Horowitz @ziebrah @david_madras Every prediction can be explained. This includes: - Predictions that cannot be changed - Predictions that are unfair - Predictions that are uncertain Explaining these predictions does more harm than good.

0

2

6

Berk Ustun

@berkustun

6 years

@mikarv @m_sendhil In terms of the setting considered in the paper, my opinion is that simpler models shouldn’t be ruled out b/c they can be more easily understood and contested by more people.

1

0

6

Berk Ustun

@berkustun

6 years

@ruchowdh @KayFButterfield @johnchavens @vdignum @jonniepenn @j2bryson @tabithagold @maria_axente @robmccargow We should try to use “justice” instead of “fairness” to make it clear that we’re not talking about “parity.” There’s a nice story about this in the talk by @Debbie_Hellman at @fatconference last year.

FAT* 2018: Keynote by Deborah Hellman - What is discrimination, when...

Speaker: Deborah Hellman, University of Virginia School of Law, D. Lurton Massee Professor of Law, Roy L. and Rosamond Woodruff Morgan Professor of LawDiscus...

www.youtube.com

1

0

5

Berk Ustun

@berkustun

6 years

“Grungy boots-on-ground work is how we build our intuitions about what kinds of solutions actually work vs. sounding good on paper. It is hard — though not impossible — to skip this step and still do great work.” @johnregehr

0

5

Berk Ustun

@berkustun

5 years

@_beenkim It’s arguably the best tool that researchers have to share their opinions with others. It’s much easier and way more effective to tweet about something than write a blog post or position paper.

1

0

5

Berk Ustun

@berkustun

4 years

@yuvalmarton @CharlieTMarx @ShlomoArgamon @FlavioCalmon @emilymbender I thought it was implied, but I think you're right – we'll update the arXiv paper with a statement to clarify our position.

1

0

5

Berk Ustun

@berkustun

6 years

ML PSA: Stop using 20% of the data as an independent test set. Train your ML model with all the data. Use 5-CV to pair this model with an estimate of predictive performance. If method needs parameter tuning, then use nested CV to avoid bias. More info here

0

1

4

Berk Ustun

@berkustun

6 years

@geomblog Hmm interesting. To be clear, is the argument: it might be possible (possibly easier) to first train a black-box, and then distill this into a white-box by producing an explanation that has 100% fidelity over all inputs?

1

0

5

Berk Ustun

@berkustun

5 years

On my reading list...

Vivian Lai

@vivwylai

5 years

(1/n) Here's a tweet thread about our paper accepted at @fatconference ! Link to paper: . Our goal is to understand how we can use machine learning models to enhance human decision making while retaining human agency.

1

4

36

0

5

Berk Ustun

@berkustun

6 years

Last week the US Dept of HUD filed a discrimination lawsuit against Facebook: "When FB uses the vast amount of personal data it collects to help advertisers to discriminate, it's the same as slamming the door in someone's face."

1

2

5

Berk Ustun

@berkustun

5 years

@Aaroth @tmiller_unimelb @adrian_weller @lilianedwards @mikarv Some topics on the frontiers: Human Subjects Experiments Adversarial Interpretations & Fairwashing Multiplicity of Good Models

"How do I fool you?": Manipulating User Trust via...

As machine learning black boxes are increasingly being deployed in critical domains such as healthcare and criminal justice, there has been a growing emphasis on developing techniques for...

arxiv.org

2

1

5

Berk Ustun

@berkustun

5 years

@Aaroth @tmiller_unimelb 3. Transparency: Motivations and Challenges by @adrian_weller 4. Slave to the Algorithm by @lilianedwards @mikarv 2/2

1

0

4

Berk Ustun

@berkustun

4 years

@alexdamour It seems to be working?!

1

0

5

Berk Ustun

@berkustun

6 years

“A survey of 400 algorithms presented in papers at two top AI conferences in the past few years… found that only 6% of the presenters shared the algorithm’s code” Missing data hinder replication of artificial intelligence studies.

0

1

5

Berk Ustun

@berkustun

5 years

ML folks — have you worked with datasets with noisy or corrupted labels? If so, please share! Trying to get a better sense of applications where they crop up & how people handle it in practice.

5

2

5

Berk Ustun

@berkustun

6 years

GDPR PSA: The “logic” of an SVM model is to maximize the margin on the training data. Generating an “explanation” for this model leads to misleading rationalizations. There are many such explanations, and the explanations are only valid for the training dataset. #FAT2018

0

5