Shashwat Goel @ShashwatGoel7 Twitter profile | Pikagi

Pikagi

Shashwat Goel

@ShashwatGoel7

424

Followers

361

Following

17

Media

126

Statuses

Researcher, currently thinking about the Science of Data and Evaluations in Deep Learning. Past: @iiit_hyderabad , @exunclan , @dpsrkpnet

New Delhi, India

https://t.co/z8dcY6Kmon

Joined June 2020

Don't wanna be here? Send us removal request.

Pinned Tweet

@ShashwatGoel7

Shashwat Goel

4 months

Work with @nikhilchandak29 , @DominikPeters just won the Outstanding Paper Award🏆at AAAI 2024 (top-3/12000+). We study a framework for making multiple decisions that fairly satisfy diverse preferences. Potential applications range from faculty hiring to how AI combines values🧵👇

Tweet media one

10

14

179

Last Seen Profiles

@clever7k

@grimkastle

@grindincoffeeco

@RosalieDutheXXX

@yaman_kosf

@michiko36358200

@michelesw7

@L0V3TAYL0RSW1FT

@jandakembangstw

@namednegin

@psicojackie_gte

@Drum_United

@MYuanchao

@shanixbhatti

@useAerial

@spookzerss

@055Pwc

@Misha222_

@aeyakovenko

@bebadafavorita

@wormwifetwitch

@not19fin

@Sun_Gooner

@ssupmiks

@xSnoopDog_

@nao419530418676

@swiftielm

@VikingNAFOFella

@iiiobk

@12_Riiin

@is__v

@5757Taka

@FlyWithInk

@hekira123

@kikaipenguin

@CattaniMauricio

@ShashwatGoel7

Shashwat Goel

10 days

Students at @iiit_hyderabad , supposed to be one of the top engineering colleges in the country, are in an ongoing health crisis caused due to appalling mismanagement and negligence going on since well over a year. 🧵👇on mass typhoid breakouts, food poisoning, underreporting..

77

492

3K

@ShashwatGoel7

Shashwat Goel

10 days

Students are forced to subscribe to the college 'mess' (apt word). Cockroaches in food, flies, lack of handwash etc. are just meant to be ignored, since years. The fact that there's less food than oil is somehow not even a major concern. Student complaints are ignored.

Tweet media one

5

45

340

@ShashwatGoel7

Shashwat Goel

10 days

When students report sickness, hostel and health authorities blame it on food orders from @swiggy @zomato , which frankly are much safer. It's so useless that students have given up and stopped even trying to report. Claiming plausible deniability is unfortunately a trend.

1

21

326

@ShashwatGoel7

Shashwat Goel

10 days

Luckily, this cover blew apart recently when @OlympiadPanini high-school students got severely sick and hospitalised, and they weren't ordering food. The worst part is, the college hid this from the rest of the student body, leading to dozens of avoidable sick students.

1

27

288

@ShashwatGoel7

Shashwat Goel

10 days

Being sick every month is just a part of Life @IIIT . For an institute that's proud of its CS research output, maybe it's time to look inward at the living conditions of the students driving this research. Won't even get started on codified exploitation in research practices.

1

25

224

@ShashwatGoel7

Shashwat Goel

10 days

I'm writing this here only because current students can't due to repercussions, and alumni are just relieved their time here is over. Maybe this stops faculty from turning a blind eye. My hunch is this is common in many top engineering colleges in 🇮🇳, but it really shouldn't be.

4

21

227

@ShashwatGoel7

Shashwat Goel

10 days

This isn't the first time. As @pingiiit reported, last year there was a widespread Typhoid breakout with 40+ cases due to contaminated water. The boys hostel warden intimidated students from getting tested, actively spreading false information about symptoms, worsening things.

Tweet media one

1

27

217

@ShashwatGoel7

Shashwat Goel

10 days

This was met with coverups, scientific falsehoods and false promises, which often got exposed on internal mailing lists. New water coolers were installed, mostly in academic and research buildings, not changing the OBH 3rd floor cooler which caused most cases.

1

15

206

@ShashwatGoel7

Shashwat Goel

10 days

For some more (Trigger alert: insects) pictures, checkout @FoodIIITHyd

4

7

139

@ShashwatGoel7

Shashwat Goel

4 months

You find some training data sources were compromised, potentially causing #backdoors , #bias , #mislabels etc. Can you remove its influence from previously trained #ML models instead of stopping their use? New work with @AmyPrb @AmartyaSanyal @ponguru @OxfordTVG studies this🧵👇

Tweet media one

2

13

43

@ShashwatGoel7

Shashwat Goel

2 months

When @sinha_shiven told me he wants to work on AI solving IMO problems with Indian univ compute I was skeptical... Just a month later, so glad to be surprised! They almost matched AlphaGeometry with 0 GPUs, within 5 minutes. Questions the hype around LLMs for solving math

@sinha_shiven

Shiven Sinha

2 months

Excited to announce our preprint! We develop a symbolic system for IMO Geometry that can rival Silver Medalists. Combined with AlphaGeometry, it outperforms IMO Gold Medalists in Geometry for the first time 🏅

Tweet media one

8

73

348

0

3

21

@ShashwatGoel7

Shashwat Goel

9 days

@alltimecoder @iiit_hyderabad So note that despite this situation, students are still being forced to pay for the mess, even if they don't eat there.

1

1

37

@ShashwatGoel7

Shashwat Goel

24 days

Starting my public thesis presentation/defence in 2 minutes. Scan the QR code in the poster to join if bored :)

@ponguru

Ponnurangam Kumaraguru “PK”

1 month

@ShashwatGoel7 's #MSThesisDefense 🗣️🗣️ "New Frontiers for Machine Unlearning" 1500hrs IST, 23rd May, IIITH. #OpenToAll #AnybodyCanAttend #PrecogsRock #ProfGiri #Student29 📜Full thesis: S's papers: {DMLR at ICLR, ICML, AAAI} 2024 & RepL4NLP at ACL 2023

Tweet media one

0

1

7

1

0

15

@ShashwatGoel7

Shashwat Goel

2 months

Will be attending @iclr_conf and the @DMLRWorkshop in Vienna. Would love to chat about all things Alignment, Interpretability, Data-Centric AI, and how models (should) deal with conflicting training data, or any application that excites you :) DM/Reply #ICLR #ICLR2024

@ponguru

Ponnurangam Kumaraguru “PK”

3 months

📢 🎉 Our📚 Corrective #Machine #Unlearning paper accepted at DMLR @ICLR '24 2024 w/ @ShashwatGoel7 @AmyPrb @AmartyaSanyal @OxfordTVG #compromised #data #backdoors #bias #mislabels #ML #models 📃 💻 #AcademicTwitter 🧵👇

Tweet media one

1

5

33

0

2

14

@ShashwatGoel7

Shashwat Goel

1 month

Reviewing recognitions are always satisfying 😌

@DMLRWorkshop

Workshop on Data-centric Machine Learning Research

2 months

We extend our appreciation to the exceptional reviewers whose expertise enabled a high-quality selection process for #DMLRWorkshop @iclr_conf 2024! Magdalena Proszewska Yifan Zhang Miguel de Benito Delgado Agam Goyal Shashwat Goel David Esiobu Ben Feuer Ian Beaver

0

0

2

0

1

11

@ShashwatGoel7

Shashwat Goel

3 months

One of the things IIIT-H gets right

@xennygrimmato_

Vaibhav Tulsyan

@xennygrimmato_

3 months

I’ve spent some time studying Russian and Chinese school programming culture. The obvious: Do what China and Russia do - incentivise college admissions based on results. Not so obvious: Create a better supply of teachers who can guide young students and scout top talent early on.

3

3

57

0

0

11

@ShashwatGoel7

Shashwat Goel

3 months

We demonstrate the use of unlearning to remove potentially harmful dual-use knowledge and capabilities from LLMs. Was super cool seeing this play out from a small exploratory project during my time at @MATSprogram to such a large scale collaboration! See the TIME article link👇

@ai_risks

Center for AI Safety

3 months

The White House Executive Order on AI highlights the risks of LLMs empowering malicious actors in developing biological, cyber, and chemical weapons. To measure and reduce these risks, we’re releasing the Weapons of Mass Destruction Proxy (WMDP) benchmark. (🧵below)

Tweet media one

4

22

58

1

1

11

@ShashwatGoel7

Shashwat Goel

18 days

Surprising climate change is a major political issue everywhere in the world, but not one in India, despite the national capital facing some of its worst effects.

@spectatorindex

The Spectator Index

@spectatorindex

18 days

BREAKING: 🇮🇳 Delhi reports record high temperature of 52.3 degrees Celsius

386

2K

8K

0

0

11

@ShashwatGoel7

Shashwat Goel

18 days

Asked GPT-4o to generate MCQ questions with answers from a blog for a quiz I'm making. 50%+ answers are c), and 80%+ are either b) or c). The ratio of b/c increases as I ask it to make more challenging questions. Possibly reveals something about the internet distribution of mcqs?

2

0

11

@ShashwatGoel7

Shashwat Goel

1 month

Great introduction to the basic ideas of Unlearning, collating the many perspectives in the field

@kenziyuliu

Ken Liu

1 month

The idea of "machine unlearning" is getting attention lately. Been thinking a lot about it recently and decided to write a long post: 📰 Unlearning is no longer just about privacy and right-to-be-forgotten since foundation models. I hope to give a gentle

22

160

740

0

1

10

@ShashwatGoel7

Shashwat Goel

2 months

Things I never expected: @AmyPrb becoming hot in the pro-GOFAI community. Follow him for a very cool GOFAI result coming out soon. Spoiler: GOFAI matches AlphaGeometry with only a few minutes of CPU time

@rao2z

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)

2 months

Oh no! LLaMAI under attack.. 😱 "multimodal models require exponentially more data to achieve linear improvements in downstream “zero-shot” performance" So what if it is "exponentially more data"? We know offline data or compute complexity doesn't matter 🙄.. c.f.

Tweet media one

3

11

55

1

1

8

@ShashwatGoel7

Shashwat Goel

2 months

What some of these teams have achieved in just a course project is pretty impressive! Expecting multiple workshop submissions in the next few weeks

@ponguru

Ponnurangam Kumaraguru “PK”

2 months

🧑🏽‍🏫 👨🏽‍🏫Course project (list 👇🏽) poster presentation. CS7.405: Responsible & Safe AI Systems @iiit_hyderabad . Please join if this is of interest to you. Open to public / outside campus also. Course materials: #ProfGiri #RAISpring2024

Tweet media one

0

1

11

0

1

7

@ShashwatGoel7

Shashwat Goel

3 months

@StephenLCasper Identical to Figure 1 in our paper, just a level or 2 higher in abstraction

Tweet card media

Corrective Machine Unlearning

Machine Learning models increasingly face data integrity challenges due to the use of large-scale training datasets drawn from the internet. We study what model developers can do if they detect...

1

0

7

@ShashwatGoel7

Shashwat Goel

1 month

@computer_phile now has a dedicated video on @AmyPrb 's (and collaborators') recent paper. Massive flex IMO

Tweet media one

1

2

5

@ShashwatGoel7

Shashwat Goel

2 years

Thrilled to present my latest work on evaluating approaches for Machine #Unlearning -- data removal from trained ML models. Was exciting to leverage insights from Empirical DL Theory and Attacks on ML. Glad to be working with my amazing co-authors @ponguru , Ameya Prabhu.

@ponguru

Ponnurangam Kumaraguru “PK”

2 years

Our Recent Paper: Evaluating Inexact #Unlearning Requires Revisiting #Forgetting 📜 Work w/ @ShashwatGoel7 & Ameya Prabhu #MachineLearning #DeepLearning #classification #data #privacy #gdpr #bias #AISafety #scaling #AcademicTwitter Thread 🧵⬇️

Tweet media one

3

4

14

1

0

4

@ShashwatGoel7

Shashwat Goel

4 months

@furongh Our paper: may be of interest, we discuss how different preferences can be aggregated for proportional representation to different groups. This ensures decisions don't overweight contrarian individuals, something maximin can suffer from.

Tweet card media

Proportional Aggregation of Preferences for Sequential Decision Making

We study the problem of fair sequential decision making given voter preferences. In each round, a decision rule must choose a decision from a set of alternatives where each voter reports which of...

0

0

5

@ShashwatGoel7

Shashwat Goel

4 months

The same set of voters may keep on being satisfied. Some voters might end up approving no decision💔. Instead, if 30% voters agree (approve a common alternative) in 10 rounds, we want them to approve 3 or more decisions. Ideally, proportional influence even on worst-case inputs.

Tweet media one

1

1

5

@ShashwatGoel7

Shashwat Goel

4 months

Suppose you wish to decide where to hangout with your group of friends. In each 'round', every 'voter' approves (👍/👎) some 'alternatives', and one alternative is to be picked as a 'decision'. Simple approach? In each round, pick the alternative with most approvals. Whats bad?🛑

Tweet media one

1

0

4

@ShashwatGoel7

Shashwat Goel

22 days

Really cool study w neat ideas

@giffmana

Lucas Beyer (bl16)

23 days

PSA: Stop pretraining your VLMs on EN-filtered data, even if it improves ImageNet and COCO‼️ Doing so impairs the model's understanding of non-English cultures❗️ I argued for years, now finally publish concrete results for this (imo) intuitively obvious recommendation A🧾🧶

8

30

233

1

0

3

@ShashwatGoel7

Shashwat Goel

4 months

Unfortunately @nikhilchandak29 's Canada visa was not provided on time. Thankfully, Nicholas Teh () from Oxford graciously agreed to deliver the talk on Friday 2PM at Room 211 at AAAI on our behalf🙏. Drop by to learn more, and feel free to reach out to us!

1

0

3

@ShashwatGoel7

Shashwat Goel

4 months

For more details about our work, check out our video and paper. Work started during an internship at LAMSADE, Université Paris Dauphine-PSL. Special thanks to Jérôme Lang for inviting us and advising the project! 📹 📃

Tweet card media

Proportional Aggregation of Preferences for Sequential Decision Making

We study the problem of fair sequential decision making given voter preferences. In each round, a decision rule must choose a decision from a set of alternatives where each voter reports which of...

0

0

3

@ShashwatGoel7

Shashwat Goel

4 months

@AmyPrb @PJNarayanan @iiit_hyderabad Thanks for the very kind words @AmyPrb , and being one of my biggest inspirations in enjoying research! <3

0

0

3

@ShashwatGoel7

Shashwat Goel

2 years

Pleased to be recognized as an Outstanding Reviewer (Top 10%) by @icmlconf in my first attempt at reviewing! Grateful for getting the opportunity as an undergrad considering I hadn't published at an ML venue before. #ICML2022

0

0

3

@ShashwatGoel7

Shashwat Goel

2 months

@michael_nielsen Some limitations of doing this:

Tweet card media

Questioning the Survey Responses of Large Language Models

As large language models increase in capability, researchers have started to conduct surveys of all kinds on these models in order to investigate the population represented by their responses. In...

0

0

2

@ShashwatGoel7

Shashwat Goel

19 days

@StephenLCasper Strongly believe in 2-7 too. I'm uncertain about quantities for 1, how do you approximate these numbers?

1

0

2

@ShashwatGoel7

Shashwat Goel

4 months

We show an interesting implication for AI: Typical setup (eg: RLHF) of combining data from different groups and maximizing accuracy can lead to unfair outcomes, even for balanced datasets❌ Learning separate models for each group 🇮🇳🇫🇷🇺🇸 and aggregating with our rules does better!

Tweet media one

1

0

2

@ShashwatGoel7

Shashwat Goel

2 months

@_akhaliq explores drawbacks of maximizing utility from approval votes (it is highly majoritarian and ignores smaller groups), and presents ways to aggregate preferences with worst case guarantees on representation.

Tweet card media

Proportional Aggregation of Preferences for Sequential Decision Making

We study the problem of fair sequential decision making given voter preferences. In each round, a decision rule must choose a decision from a set of alternatives where each voter reports which of...

0

1

2

@ShashwatGoel7

Shashwat Goel

4 months

Our work opens many interesting directions for future work. Theoretically, can we extend from approvals to cardinal utilities? Applications: Democratic processes🧑‍⚖️; AI pursuing a mixture of (sometimes conflicting) goals 🤖; Aligning with subjective values 👨‍👩‍👧‍👧🇺🇳; and many more!

Tweet media one

1

0

2

@ShashwatGoel7

Shashwat Goel

2 years

AI Safety researcher: "Let me come up with the most intricate pathways for AGI catastrophe to show how effective my research is" AGI: "Sounds like a plan *rubs hands*", The model is probably learning to he unsafe from AI Safety posts in the training data. Ironic.

@zswitten

Zack Witten

2 years

Finally, I had to try out the paperclip test, since it's practically the Hello World of alignment at this point. Nice to know there will be a few humans left over!

Tweet media one

Tweet media two

40

311

2K

0

0

2

@ShashwatGoel7

Shashwat Goel

9 days

@alltimecoder @iiit_hyderabad They still have limits, might be more relaxed than 5/month.

0

0

2

@ShashwatGoel7

Shashwat Goel

4 months

@S__Schoepf @AmyPrb @AmartyaSanyal @ponguru @OxfordTVG That's good to know! In an earlier work we had also tried random error correction and it's nice to see corrective applications of unlearning gaining more attention! :)

Tweet card media

Towards Adversarial Evaluations for Inexact Machine Unlearning

Machine Learning models face increased concerns regarding the storage of personal user data and adverse impacts of corrupted data like backdoors or systematic bias. Machine Unlearning can address...

0

0

1

@ShashwatGoel7

Shashwat Goel

4 months

@ravi_iitm @RealAAAI @iiit_hyderabad Thank you Prof. @ravi_iitm 🙏

0

0

1

@ShashwatGoel7

Shashwat Goel

9 months

@shrimpdoll_ @akshitkr Lmao who tf said anything about iiit?!

1

0

1

@ShashwatGoel7

Shashwat Goel

18 days

@gaur_manu Saw this, but it focuses more on answering MCQ questions rather than generation. Would be interesting to see if question generation is biased towards different options from question answering

0

0

1

@ShashwatGoel7

Shashwat Goel

4 months

We believe our work highlights massive scope for new research contributions, including new evaluations🧪, unlearning methods 🔨, and theoretical analysis 📝of the Corrective Unlearning setting. For more details: 📒 👨‍💻

Tweet media one

1

0

1

@ShashwatGoel7

Shashwat Goel

23 days

@savvyRL And yes, one of these was at ICML 2023, before I had a top-tier submission. It was a great learning experience, and Ig I didn't do a bad job? Or maybe the recognitions are really noisy too.

0

0

1

@ShashwatGoel7

Shashwat Goel

1 month

@dhruvtyagiii1 @pratikpoddar Mine is voice to voice :)

1

0

1

@ShashwatGoel7

Shashwat Goel

1 year

Even more fun when your evaluation is weak or flawed. Underwater ideas seem deceptively cool.

@drjwrae

Jack Rae

1 year

Almost all research ideas work when your baseline is weak. A stronger baseline, like a rising tide, pulls a lot of them underwater.

2

1

64

0

0

1

@ShashwatGoel7

Shashwat Goel

4 months

We study popular voting rules in 3 settings: online, semi-online (only number of rounds known apriori) and offline. We provide comprehensive results on which properties each rule satisfies✅, and reduce to a well known open problem❓in most remaining cases.

Tweet media one

1

0

1

@ShashwatGoel7

Shashwat Goel

2 years

One of those where you don't realize you need it until you see it. Now search results will never satisfy me (until, maybe?)

@RichardSocher

Richard Socher

2 years

I've worked on academic deep learning and summarization for years. Summarization is a foundational technology for the information age and a remedy for the attention economy. Here's a🧵 for how we think and apply summarization at @YouSearchEngine

13

77

564

0

0

1

@ShashwatGoel7

Shashwat Goel

23 days

@savvyRL DMed. Tysm for doing this! Since there's backlash just thought I'd add my pov. I love reviewing. In the few times I got a chance, I've been recognised as an outstanding reviewer whenever there was one. I hope gatekeeping is not the solution for some of the very valid concerns :(

1

0

1

@ShashwatGoel7

Shashwat Goel

1 month

@kenziyuliu Wanted to write something like this, and you've done an excellent version of what I had in mind! Really useful resource for people starting out in Unlearning, will share widely :)

1

0

1

@ShashwatGoel7

Shashwat Goel

1 month

@DominikPeters Maybe it uses GPT4 supervision of some form? Could be in data, distillation or more

0

0

1

@ShashwatGoel7

Shashwat Goel

4 months

@monojitchou @nikhilchandak29 @DominikPeters Thank you sir! Means a lot coming from you

0

0

1

@ShashwatGoel7

Shashwat Goel

4 months

@saujasv @nikhilchandak29 Thank you for the constant support @saujasv ❤️

0

0

1

@ShashwatGoel7

Shashwat Goel

4 months

@eyal_eg @PJNarayanan @iiit_hyderabad You can find the Twitter thread, which also has the paper and video, pinned on my profile

0

0

1

@ShashwatGoel7

Shashwat Goel

8 months

@boknilev I agree that probing is similar to representation reading and more citations to this work should be added. However, the unique takeaway is the extent to which using methods like PCA/mean difference (Figure 12) finds activation directions that can control generation/removal.

0

0

0

@ShashwatGoel7

Shashwat Goel

4 months

Specifically for removing the BadNet poison, only one method (SSD) studied succeeds, showing the tractability of generalizing removal from a representative subset. However, SSD hurts model utility, leading to significant drops in test-accuracy on clean samples.

Tweet media one

1

0

1

@ShashwatGoel7

Shashwat Goel

4 months

We formalize this by adapting Justified Representation (PJR, EJR) axioms from Social Choice Theory literature. We prove our axioms are tight: solutions that satisfy stronger guarantees may not exist, and even when they exist cannot always be found online😓.

Tweet media one

1

0

1