Aditya Grover @adityagrover_ Twitter profile

Pinned Tweet

Aditya Grover

6 months

Incredibly honored to be in the #ForbesUnder30 list for 2024! Thanks to my incredible support team over the years - students, mentors, colleagues, friends, and family.

51

14

420

Last Seen Profiles

@BDKDY99

@ThelbySommy

@RobertaByTheSea

@saaachimehta

@pannone25433

@DubaiRacingTV

@philia_mp3

@YayAya529

@DustySpaceMouse

@JanetAl41140264

@HammondCoach96

@SenadM55350

@MuneguLive

@ShootingWith

@TrpsAngl

@MariaWilso78787

@jmorris_83

@Nappi2022

@mastconfex

@NoShorty_

@ripdawg99

@HariomM13299590

@DomaLatrice

@DiablesItaca

@eatmyass4lunch

@ELECroyalties

@kirstiealley

@Stetesh114142

@bamhanni

@SystematicJuice

@Paul87753264

@ChrisGreece48

@ChrisKenny76582

@maddiemartin24

@MyGoldenChips

@leesh222

Aditya Grover

@adityagrover_

1 year

📢Introducing ClimateLearn, a new PyTorch library for accessing climate datasets, state-of-the-art ML models, and high quality training and visualization pipelines. Blog: Docs: Quickstart Colab: 🧵 (1/n)

NeurIPS2022_CCAI_Tutorial.ipynb

Colaboratory notebook

colab.research.google.com

17

277

1K

Aditya Grover

@adityagrover_

4 years

With my thesis formally approved @Stanford , I am excited to share next steps: I'll be joining @CS_UCLA @UCLAengineering as an assistant professor of CS in Fall 2021. For this coming year, I'll be a researcher at @facebookai and @pabbeel 's group at @berkeley_ai .

61

17

1K

Aditya Grover

@adityagrover_

3 years

Thrilled to share that my PhD dissertation won the ACM SIGKDD Dissertation Award for "outstanding work in data science and machine learning". Thanks to everyone involved, especially my advisor @StefanoErmon & @StanfordAILab !

56

45

800

Aditya Grover

@adityagrover_

5 years

Exciting first day co-teaching CS 236: Deep Generative Models with @ermonste at Stanford! Course enrollments more than doubled from our previous iteration. All our course materials will be available at:

6

94

468

Aditya Grover

@adityagrover_

4 years

Great time presenting a tutorial on Deep Generative Models at the Deep Learning for Science School () at Berkeley! Video: Slides:

6

77

440

Aditya Grover

@adityagrover_

2 years

I am hiring PhD students in ML this cycle! Topics include unsupervised ML, generative models, seq. decision making, and climate change. For those interested, plz apply to UCLA CS. Deadline: Dec 15. Plz share widely! More info: #AI #ML #AcademicTwitter

4

85

424

Aditya Grover

@adityagrover_

4 years

And it’s a wrap to my PhD @StanfordAILab ! Thanks to my wonderful committee — @StefanoErmon , @erichorvitz , @jure , @percyliang , Moses Charikar & Stephen Boyd — for putting up a manageable offense during my defense and everyone else who has helped me along the way 🙂

Stefano Ermon

@StefanoErmon

4 years

Super proud of my student Aditya who successfully defended his #PhD dissertation today! He has done awesome work on unsupervised learning with generative models. Congrats, Dr. @adityagrover_ 👏🎊🎉

36

23

605

23

6

419

Aditya Grover

@adityagrover_

2 years

Our paper on Moser Flows got the outstanding paper award at #NeurIPS2021 !! Congratulations and thanks to all my wonderful co-authors: Noam Rozen, @mnick , @lipmanya :) If you are attending #NeurIPS2021 , please attend our oral presentation and live Q&A on Dec 10 at 4 PST.

Yaron Lipman

@lipmanya

2 years

Excited to share that our paper "Moser Flow: Divergence-based Generative Modeling on Manifolds" won an Outstanding Paper Award at NeurIPS 2021!! Noam Rozen @adityagrover_ @mnick See thread for a short summary 👇

6

43

252

11

12

247

Aditya Grover

@adityagrover_

3 years

Every PhD application season, we see many familiar questions pop up repeatedly. @hima_lakkaraju and I hosted a panel last year with many distinguished ML faculty answering many such burning questions from prospective candidates: #AI #ML #AcademicTwitter

Demystifying ML PhD Admissions to US Universities

This video is a recording of the panel discussion and Q & A on ML PhD admissions to US universities. This event happened on November 21st at 12pm ET over zoo...

www.youtube.com

3

43

220

Aditya Grover

@adityagrover_

2 years

Every PhD application season, we see many familiar questions pop up repeatedly. @hima_lakkaraju and I hosted a panel where prominent AI faculty share a behind-the-scenes view with prospective candidates: #AI #ML #AcademicTwitter

Demystifying ML PhD Admissions to US Universities

This video is a recording of the panel discussion and Q & A on ML PhD admissions to US universities. This event happened on November 21st at 12pm ET over zoo...

www.youtube.com

3

46

182

Aditya Grover

@adityagrover_

4 years

Great summary article in WIRED covering our research in using ML for accelerating battery development! Fun fact: The largest public datasets for battery research today are at least 100x larger in size than the ones from 5 years ago.

WIRED

@WIRED

4 years

Battery storage is key to increasing the amount of renewable energy on the grid, and when it comes to decarbonizing our energy supply, time is of the essence. After decades of plodding progress, AI-driven battery research promises to pick up the pace.

1

31

93

2

19

165

Aditya Grover

@adityagrover_

3 years

The recording for the panel discussion is up on Youtube: Many gem insights here for both prospective and current PhD students!

𝙷𝚒𝚖𝚊 𝙻𝚊𝚔𝚔𝚊𝚛𝚊𝚓𝚞

@hima_lakkaraju

4 years

We are organizing a panel discussion/Q&A on “Demystifying ML PhD Applications to US Universities” with faculty spanning various universities this Saturday at 9am PT/12pm ET. Please register at to attend. #AI #ML #AcademicTwitter [1/n]

8

121

447

1

15

144

Aditya Grover

@adityagrover_

2 years

📢New paper out: We show that transformers can be repurposed for uncertainty-aware learning and decision making, smashing all previous benchmarks for Neural Processes++. w/ @tungnd_13 -- first paper by my first PhD student! To appear at @icmlconf 2022.

Transformer Neural Processes: Uncertainty-Aware Meta Learning Via...

Neural Processes (NPs) are a popular class of approaches for meta-learning. Similar to Gaussian Processes (GPs), NPs define distributions over functions and can estimate uncertainty in their...

arxiv.org

Tung Nguyen

@tungnd_13

2 years

Transformers show excellent capabilities in few-shot/meta learning, but have been mostly evaluated on accuracy-based metrics. How can we represent uncertainty in meta learning with transformers? We address this question in our new work at #ICML2022 !

3

24

162

0

16

126

Aditya Grover

@adityagrover_

1 year

I am looking to hire a postdoc at UCLA in artificial intelligence and climate science. Details below. Please share and rt in your circles!

2

36

118

Aditya Grover

@adityagrover_

3 years

Very excited to share our new paper "Pretrained Transformers as Universal Computation Engines" led by the amazing @_kevinlu ! We show how to transfer pretrained language transformer to an arbitrary modality (vision, proteins etc) without any expensive finetuning. Details 👇

Igor Mordatch

@IMordatch

3 years

What are the limits to the generalization of large pretrained transformer models? We find minimal fine-tuning (~0.1% of params) performs as well as training from scratch on a completely new modality! with @_kevinlu , @adityagrover_ , @pabbeel paper: 1/8

4

71

354

0

14

117

Aditya Grover

@adityagrover_

5 months

Very excited to release Stormer, the new SOTA for ML-based weather forecasting. Personally, this reinforces the bitter lesson - scale and simplicity extends to AI for science. Great collab b/w @UCLA & @argonne led by @tungnd_13 . Look fwd to chatting more at #NeurIPS2023 .

Tung Nguyen

@tungnd_13

5 months

Introducing Stormer, a scalable transformer model for skillful and reliable medium-range weather forecasting. Stormer achieves competitive accuracy for short-range, 1–7 day forecasts, while outperforming Pangu-Weather and Graphcast by a large margin for longer lead times. Paper:

10

83

475

3

11

114

Aditya Grover

@adityagrover_

10 months

After a successful presentation at the #ICML2023 main conference, delighted to share that ClimaX was also recognized with a Best Paper Award at the ICML Workshop on Scientific and ML modeling! Congrats to all coauthors @tungnd_13 @jo_brandstetter @akapoor_av8r @rejuvyesh !

Aditya Grover

@adityagrover_

1 year

ClimaX will appear in #ICML2023 ! 🏖️🧑‍💻🌎 Bonus: @tungnd_13 will also be giving a spotlight talk next week at the upcoming #ICLR2023 workshop on Tackling Climate Change with AI.

1

8

42

4

11

114

Aditya Grover

@adityagrover_

2 years

One of our key motivations with Decision Transformers was to bring closer the research communities in sequential decision making, language, and perception. Thank you @huggingface for the integration! 🤗

Thomas Simonini ᯅ

@ThomasSimonini

2 years

We integrated Decision Transformers, an Offline Reinforcement Learning method, into the 🤗 transformers library and the @huggingface Hub 🥳 ➕ 9 pre-trained models for continuous control tasks in Gym 🔥 We wrote a tutorial if you want to try it 👉

5

61

240

0

11

111

Aditya Grover

@adityagrover_

5 years

Our @iclr2019 paper proposes NeuralSort, a differentiable relaxation to sorting. Bonus: new Gumbel reparameterization trick for distributions over permutations. Check out our poster today at 4:30! Code: w/ @tl0en A. Zweig @ermonste

2

18

109

Aditya Grover

@adityagrover_

5 months

Great time speaking at the IndoML conference at IIT Bombay about the latest research from my group on in-context learning and my thoughts on the future of generative AI. Thank you to the organizers for inviting me, and bringing together bright young minds from all over India.

IndoML Symposium, 2023

@indoml_sym

5 months

Day 1: Session 1 Foundational and Generative Models Talk 2: Aditya Grover Professor, UCLA His three pronged talk on TNP, ExPT and GPO for few shot surrogate modelling, experimental design and group preference optimisation was really thought-provoking!

0

5

22

2

8

103

Aditya Grover

@adityagrover_

2 years

When using models such as CLIP for zero-shot predictions, how often do language and image representations agree? Our latest work ⬇️ reformulates the contrastive loss for multimodal data with significantly improved consistency and zero-shot robustness

AK

@_akhaliq

2 years

CYCLIP: Cyclic Contrastive Language-Image Pretraining abs:

1

21

155

2

12

101

Aditya Grover

@adityagrover_

1 year

What makes weather and climate any different from language or vision? A lot if you ask a physicist, but less so under the lens of today's AI. Very excited to release ClimaX! Our vision is to develop a new scientific basis for Earth systems driven by ML foundation models.

Tung Nguyen

@tungnd_13

1 year

Introducing ClimaX, the first foundation model for weather and climate. A fast and accurate one-stop AI solution for a range of atmospheric science tasks. Paper: Blog: Thread🧵 #ML #Climate #Weather #FoundationModel

35

179

851

4

10

100

Aditya Grover

@adityagrover_

2 years

To appear in #Neurips2022 ! Congrats to my students @_shashankgoel_ and @hbXNov . For those training CLIP models, check out how a simple symmetrization of the contrastive loss can boost zero shot robustness.

AK

@_akhaliq

2 years

CYCLIP: Cyclic Contrastive Language-Image Pretraining abs:

1

21

155

0

8

95

Aditya Grover

@adityagrover_

3 years

Incorporating rotation invariance in neural nets is key for predicting molecular properties such as energies. Led by Larry Zitnick, our latest work achieves this via convolutions w.r.t. a local coordinate frame. State of the art on the OC20 dataset for catalyst discovery!

arXiv Daily

@Arxiv_Daily

3 years

Rotation Invariant Graph Neural Networks using Spin Convolutions by @mshuaibii et al. #DegreeofFreedom #MachineLearning

0

6

26

0

8

91

Aditya Grover

@adityagrover_

5 months

Except that they already did...at least a year back! Quite strange how this paper ignores all prior work in diffusion models for offline RL. Diffuser (ICML-22), Diffusion Q-Learning (ICLR-23), Decision Diffuser (ICML-23), Decision Stacks (NeurIPS-23; )

Decision Stacks: Flexible Reinforcement Learning via Modular...

Reinforcement learning presents an attractive paradigm to reason about several distinct aspects of sequential decision making, such as specifying complex goals, planning future observations and...

arxiv.org

Jakob Foerster

@j_foerst

5 months

Diffusion models have revolutionised a number of areas in ML, now they are coming for offline RL. In our paper we guide the samples to be closer to our current policy, reducing the off-policy-ness of the generated data. This will unlock novel world applications of off-policy RL.

4

19

134

2

10

89

Aditya Grover

@adityagrover_

4 years

Prospective students: I will be recruiting PhD students to join my group in Fall 2021. If interested, please apply to the CS department at UCLA (deadline: Dec 1, 2020) and list my name in your application.

3

9

84

Aditya Grover

@adityagrover_

3 years

Data for many scientific domains, such as climate, materials, and biology, lies on manifolds. How do we learn generative models here? Moser's classic results (1965) + Flows to our rescue! SOTA on earth science benchmarks and (first!) scaling to curved surfaces. Details 👇

Yaron Lipman

@lipmanya

3 years

New paper: Introducing Moser Flows (MFs), a new class of continuous normalizing flows (CNFs) on manifolds based on divergences of neural nets. First generative modeling results on general curved surfaces! with Noam Rozen @adityagrover_ @mnick 1/7

3

61

297

1

11

80

Aditya Grover

@adityagrover_

4 years

Excited to co-host this panel discussion and Q&A! Details in the thread below.

𝙷𝚒𝚖𝚊 𝙻𝚊𝚔𝚔𝚊𝚛𝚊𝚓𝚞

@hima_lakkaraju

4 years

We are organizing a panel discussion/Q&A on “Demystifying ML PhD Applications to US Universities” with faculty spanning various universities this Saturday at 9am PT/12pm ET. Please register at to attend. #AI #ML #AcademicTwitter [1/n]

8

121

447

0

9

75

Aditya Grover

@adityagrover_

4 years

Improvements in battery technology are critical for building a green and sustainable planet. Co-leading this project (w/ Peter Attia) for the last 4 years of my PhD has been an absolutely amazing "learning" experience in inter-disciplinary collaboration and patience @nature

Stefano Ermon

@StefanoErmon

4 years

Want to recharge your electric vehicle in 10 minutes? Check out our @nature paper on optimizing battery charging protocols with machine learning👉 🔋 battery testing times slashed by nearly 15-fold via @Stanford

2

52

272

1

10

72

Aditya Grover

@adityagrover_

6 years

Website for our #NIPS2018 Workshop on Relational Representation Learning (R2L) is up! Workshop date: 8th Dec. Papers due on 19th Oct. @paroma_varma @fredsala #stevenholtzen @ProfJenNeville @ermonste @HazyResearch

1

34

67

Aditya Grover

@adityagrover_

5 years

List of accepted papers and pdfs for the Relational Representation Learning Workshop at #NeurIPS2018 is up! Extremely thankful to our reviewers! #R2L @paroma_varma @fredsala @zengola @ProfJenNeville @ermonste @HazyResearch

1

24

67

Aditya Grover

@adityagrover_

2 years

Honored to receive the @MetaAI research award for our proposal to use ML for efficient optimization of scientific designs. (1/2)

Announcing the winners of the 2022 AI4AI Research request for proposals | Meta Research

In August, Meta launched the 2022 AI4AI Research request for proposals (RFP). Today, we’re announcing the winners of this award.

research.facebook.com

2

3

64

Aditya Grover

@adityagrover_

5 years

In our #NeurIPS2018 paper, we propose a streamlined variational inference approach for branching during search and demonstrate its utility on some of the hardest constraint satisfaction problems. Code: w/ Tudor Achim & @ermonste

1

9

59

Aditya Grover

@adityagrover_

4 years

Never in my wildest dreams did I imagine that one day our Flow-GAN architecture () would generate a #rickroll 😂

Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in...

Adversarial learning of probabilistic models has recently emerged as a promising alternative to maximum likelihood. Implicit models such as generative adversarial networks (GAN) often generate...

arxiv.org

On Mastodon @[email protected]

@ben_j_lindsay

4 years

If you think the GPT demos with just text are impressive, check this out. I built a model that generates music videos based on text input by replacing the final layers of GPT-3 with a Flow-GAN architecture. [Sound on]

38

304

1K

3

5

56

Aditya Grover

@adityagrover_

3 years

@yudapearl Thanks so much, Judea! Your work is an inspiration to so many young scientists such as myself :)

1

0

53

Aditya Grover

@adityagrover_

4 months

. @johnamqdang has 2/2 papers accepted at #ICLR2024 exploring alignment of LLMs with human preferences. He is also looking for PhD positions, and would be a great fit at the top departments!

John Dang

@johnamqdang

4 months

🎉Happy to share 2 papers accepted to #ICLR2024 w/ @hbXNov @siyan_zhao @adityagrover_ ! 1) Peering Through Preferences: Unraveling Feedback Acquisition for Aligning LLMs: 2) Group Preference Optimization: Few-Shot Alignment of LLMs

3

8

78

1

5

50

Aditya Grover

@adityagrover_

6 years

Amidst all the interest in compressed sensing using deep generative models, our new paper shows that sparsity assumptions can co-exist & aid generative models. Long oral at @icmlconf next week! w/ Manik Dhar & @ermonste Paper: Code:

0

14

50

Aditya Grover

@adityagrover_

6 years

Check out our new @icmlconf paper on algorithms and applications for unsupervised representation learning of policies in multiagent systems: w/ @alshedivat @rejuvyesh #YuraBurda #HarrisonEdwards @OpenAI Accepted as a long talk at #ICML2018 !

0

13

45

Aditya Grover

@adityagrover_

1 year

Thank you for the kind words. Great time at @UCSanDiego interacting with the scientific ML community on ClimaX, ClimateLearn, and promising paths ahead. Thanks @yuqirose for the invitation and organizing this event!

Preethi Karpoor

@astrokarpoor

1 year

For someone who has ZERO clue on Climate or Climate Modeling, @adityagrover_ is the teacher I need! It was such a pleasure to hear about his work. Also, I’m a bit biased towards Vision Transformers having worked on it previously, so all in all what a fantastic talk!

2

1

15

0

1

46

Aditya Grover

@adityagrover_

2 years

So much exciting work in RL foundation models pretrained on large offline datasets. Beyond pretraining, how well can these models be finetuned to explore? Check out our Long Oral Presentation on Online Decision Transformers this Thu (7/21) at #ICML2022 .

Qinqing Zheng

@qqyuzu

2 years

Can sequence models balance exploration and exploitation? Excited to share our latest work proposing online decision transformers for online+offline reinforcement learning via sequence modeling! w/ @yayitsamyzhang @adityagrover_

2

23

139

1

10

46

Aditya Grover

@adityagrover_

4 years

Super grateful to my advisor @StefanoErmon , collaborators, friends, and family for their support during my entire PhD, especially while hunting for a job in the middle of a pandemic. Will miss the wonderful vibe at @StanfordAILab !

1

0

44

Aditya Grover

@adityagrover_

2 years

Agreed! Now more than ever, different fields of AI can talk/see/hear each other (pun intended). If you haven't read this yet, here is our paper on Transformers as Universal Computation Engines:

Pretrained Transformers as Universal Computation Engines

We investigate the capability of a transformer pretrained on natural language to generalize to other modalities with minimal finetuning -- in particular, without finetuning of the self-attention...

arxiv.org

Andrej Karpathy

@karpathy

2 years

👌👌👏 I wrote earlier about the ongoing consolidation in AI towards transformer architecture from a mostly practical viewpoint, but there are also major implications on paths towards AGI exactly along the lines of this post (& Transformers as Universal Computation Engines paper)

24

78

732

1

42

Aditya Grover

@adityagrover_

6 months

Very excited to share this work! Using foundation models for scientific experimentation is a vast opportunity to connect their capabilities with the physical world and accelerate scientific discoveries.

Tung Nguyen

@tungnd_13

6 months

Introducing ExPT, a general-purpose model for few-shot experimental design (ED) that combines unsupervised pretraining and in-context learning. ExPT solves challenging ED problems with only a handful of samples. To appear at #NeurIPS2023 ! Paper:

3

19

66

0

4

43

Aditya Grover

@adityagrover_

3 years

Excited to help organize this new challenge at #NeurIPS2021 ! The challenge brings together a range of research directions in graph neural nets, sequential modeling, and distribution shifts. More details:

AI at Meta

@AIatMeta

3 years

We are launching the Open Catalyst Challenge, an open AI research competition to build new machine learning models that will help scientists discover new catalysts for efficient, economical green energy storage. Learn more:

4

89

259

1

8

42

Aditya Grover

@adityagrover_

1 year

ClimaX will appear in #ICML2023 ! 🏖️🧑‍💻🌎 Bonus: @tungnd_13 will also be giving a spotlight talk next week at the upcoming #ICLR2023 workshop on Tackling Climate Change with AI.

Aditya Grover

@adityagrover_

1 year

What makes weather and climate any different from language or vision? A lot if you ask a physicist, but less so under the lens of today's AI. Very excited to release ClimaX! Our vision is to develop a new scientific basis for Earth systems driven by ML foundation models.

4

10

100

1

8

42

Aditya Grover

@adityagrover_

3 years

Excited to share our latest work rethinking RL from a sequence modeling perspective!

Igor Mordatch

@IMordatch

3 years

Can RL algorithms be replaced with transformer-based language models? We’ve looked at this question with our work on Decision Transformer: Website: Code: 1/8

18

257

1K

0

1

41

Aditya Grover

@adityagrover_

4 years

Special JSTAT issue on selected conference papers interfacing ML and physics: Includes our extended #NeurIPS2018 paper on solving constraint satisfaction problems (e.g., SAT) using variational inference (w/ @tachim @ermonste ) and many other cool works!

0

4

40

Aditya Grover

@adityagrover_

2 years

The restriction to offline settings was a key limitation of prior works in RL via autoregressive generative models (eg, GPT). Our latest work sheds light on enabling online+offline RL while retaining much of the simplicity and effectiveness of sequence modeling.

Qinqing Zheng

@qqyuzu

2 years

Can sequence models balance exploration and exploitation? Excited to share our latest work proposing online decision transformers for online+offline reinforcement learning via sequence modeling! w/ @yayitsamyzhang @adityagrover_

2

23

139

0

1

37

Aditya Grover

@adityagrover_

9 years

Think beyond state-of-the-art deep learning in NLP; a great paper by a DL pioneer, but first a linguist @chrmanning

0

21

34

Aditya Grover

@adityagrover_

2 years

Causal reasoning begins with a causal model. Our #NeurIPS2021 paper proposes Bayesian Causal Discovery (BCD) Nets - a VAE style approach for learning a distribution over causal models using only observational data. Simple and scalable, works extremely well in practice! Details 👇

Chris Cundy

@ChrisCundy

2 years

How can you scalably form a posterior over possible causal mechanisms generating some observational data? Find out in our NeurIPS poster tomorrow (Tuesday), 8.30-10am pacific time! Joint work with @StefanoErmon , @adityagrover_ (1/13)

2

5

47

1

5

35

Aditya Grover

@adityagrover_

3 years

Looking forward to hosting the research roundtable on "Graph-based and combinatorial machine learning" (Table #23 )! w/ @ProfJenNeville #Neurips2020

WiML

@WiMLworkshop

3 years

We would like to express our sincere gratitude to the inspiring cohort of #WiML2020 MENTORS! We have over 120 mentors from academia & industry! Mentorship roundtables are divided into 3 areas Research (Tables 1–28), Career & Life Advice (Tables 29–50) and Sponsor (Tables 51–63).

3

6

95

0

2

33

Aditya Grover

@adityagrover_

5 years

Our @icmlconf paper on Graphite introduces a GNN decoder for latent variable modeling of graphs based on iterative refinement of low-rank approximations. Talk on Thurs at 10 AM in Hall A and poster at 6:30 PM. w/ Aaron Zweig, @ermonste

0

10

33

Aditya Grover

@adityagrover_

11 months

Introducing Decision Stacks: A new foundation model for RL ✅Modular design for flexible use of generative models (transformer, diffusers, etc) ✅ AR chaining of modules for maximal expressivity ✅Simple and v. effective! Beats all previous transformer, diffusion policies 🧵👇

Siyan Zhao

@siyan_zhao

11 months

Modularity is critical for design of successful software and AI systems🧩. How can we apply this principle to foundation models for Reinforcement Learning? Our latest work, Decision Stacks, improves flexibility of RL through modular generative models. w/ @adityagrover_ 🧵👇(1/8)

3

25

118

0

4

33

Aditya Grover

@adityagrover_

2 years

Excited to be co-organizing the first AI for Climate Science Bridge Program at #AAA23 ! Submit your papers by Nov 18. Further details here: w/ @rejuvyesh @akapoor_av8r @manmeet3591 @niyogidev cc: @RealAAAI @ClimateChangeAI

AAAI

@RealAAAI

2 years

AAAI seeks proposals for bridges to be held during #AAAI23 . Each bridge combines elements of education, collaboration and outreach. Bridge submissions are due to organizers on Friday, November 18. Details and deadlines:

0

1

4

0

10

32

Aditya Grover

@adityagrover_

3 years

My thesis was titled "Learning to Represent and Reason under Limited Supervision". Link: I will also be presenting my award talk at KDD on August 15 from 4-5 PM PDT.

3

31

Aditya Grover

@adityagrover_

6 months

Great work coming from Google DeepMind on medium-range weather forecasting! Thank you @Melissahei @techreview for quoting my thoughts on AI-powered forecasts steadily becoming the new normal👇

MIT Technology Review

@techreview

6 months

Google DeepMind’s weather AI can forecast extreme weather faster and more accurately

2

14

48

2

28

Aditya Grover

@adityagrover_

2 years

📢 New #ICML2022 paper: Divergences between probability distributions are well understood. How do we compare a sequence of time-indexed prob dists (e.g., in continuous normalizing flows)? We use mass conservation laws to derive new "probability path divergence" objectives 🧵👇

Heli Ben-Hamu

@helibenhamu

2 years

In our new ICML paper, we introduce CNFM: Continuous Normalizing Flow Matching. A new approach for scalable CNFs on manifolds, the first generative model to scale to moderately high dimensional manifolds! (1/n)

1

32

122

0

27

Aditya Grover

@adityagrover_

7 months

Tired of prompt engineering to make your language models work? Check out GPO, our new few-shot alignment algorithm to customize language models to preferences of different groups. w/ @siyan_zhao and @johnamqdang 🧵👇

Siyan Zhao

@siyan_zhao

7 months

How can we align LLM outputs to better represent group preferences of diverse groups🌎? We introduce Group Preference Optimization (GPO): a transformer module for few-shot alignment of LLMs to group preferences. It can adapt to unseen group preferences in a gradient-free and

2

11

80

0

4

24

Aditya Grover

@adityagrover_

1 year

Many promising results in using ML for weather & climate. But, problems abound: outdated models and non-standardized datasets and evaluation protocols. ClimateLearn reimagines the entire ML stack for easy, rigorous, and reproducible data-driven climate science. (2/n)

1

3

24

Aditya Grover

@adityagrover_

1 year

We also previewed an early version of our library at a spotlight (!) tutorial at the NeurIPS @ClimateChangeAI Workshop. Slides and recording here:

Machine Learning for Predicting Climate Extremes

Climate Change AI - NeurIPS 2022 Accepted Work

www.climatechange.ai

1

4

23

Aditya Grover

@adityagrover_

1 year

Many more features: Integrated wandb support for monitoring your ML models across a range of custom metrics. Easily produce visualizations to gain insights into model performance.

1

2

23

Aditya Grover

@adityagrover_

1 year

Our code is open-sourced at: We welcome community contributions and feedback of any kind: new datasets, models, feature requests, bug reports/fixes.

GitHub - aditya-grover/climate-learn: Source code for ClimateLearn

Source code for ClimateLearn. Contribute to aditya-grover/climate-learn development by creating an account on GitHub.

github.com

2

23

Aditya Grover

@adityagrover_

5 years

Thanks for the invitation @guyvdb and @zengola ! Great time interacting with the ML community at @UCLA about our latest #NeurIPS2019 work:

Bias Correction of Learned Generative Models using Likelihood-Free...

A learned generative model often produces biased statistics relative to the underlying data distribution. A standard technique to correct this bias is importance sampling, where samples from the...

arxiv.org

Guy Van den Broeck

@guyvdb

5 years

We have @adityagrover_ visiting #UCLA today, giving a talk about "Mitigating Bias in Generative Modeling".

0

1

31

0

1

23

Aditya Grover

@adityagrover_

5 years

@sam_sinha_ @ermonste Yes, lecture videos will be posted online in due course!

12

1

21

Aditya Grover

@adityagrover_

5 years

Turning a bug into a feature via the continuous Bernoulli distribution!

Stat.ML Papers

@StatMLPapers

5 years

The continuous Bernoulli: fixing a pervasive error in variational autoencoders. (arXiv:1907.06845v1 [])

0

2

11

0

1

21

Aditya Grover

@adityagrover_

2 years

How does exploration vs exploitation affect reward estimation? Excited to share our #AISTATS2022 paper that constructs optimal reward estimators by leveraging the demonstrator's behavior en route to optimality. 🧵:

1

20

Aditya Grover

@adityagrover_

1 year

Dataset: Load your favorite weather and climate dataset in just a few lines of code, slice it as you want. Our dataset library is growing fast, suggestions welcome!

1

2

20

Aditya Grover

@adityagrover_

1 year

Models: Train state-of-the-art deep learning models for weather forecasting, climate downscaling, climate projections. Have more models/tasks in mind? Let us know!

1

2

19

Aditya Grover

@adityagrover_

6 years

For those attending @IJCAI_ECAI_18 / @icmlconf , @ermonste and I will be giving a tutorial on deep generative models from 2-6pm today (Sat) at Hall T2. From the basics of Bayes nets to the latest Glow-ing research, everything is on the menu. Outline & Slides:

0

6

19

Aditya Grover

@adityagrover_

1 year

Excited to share 2 new works to be presented at tomorrow's @ClimateChangeAI workshop at #NeurIPS2022 : 1. Tutorial on ML for predicting climate extremes. Oral Presentation! 3:30-4pm PST w/ my students @hbXNov @_shashankgoel_ @tungnd_13 @jasonjewik (1/2)

Machine Learning for Predicting Climate Extremes

Climate Change AI - NeurIPS 2022 Accepted Work

www.climatechange.ai

Climate Change AI

@ClimateChangeAI

1 year

Our workshop, Tackling Climate Change with ML, is happening at #NeurIPS2022 next Friday, Dec 9! Keynotes about work at the intersection of #ML and #ClimateChange by Gustau Camps-Valls ( @isp_uv_es ), @inesliaz and @yuqirose . Find out more at:

0

15

50

1

4

19

Aditya Grover

@adityagrover_

6 years

How do we accelerate scientific research and discovery when experiments are time-intensive? Check out our #AISTATS2018 paper on experimental design with delayed feedback w/ @ermonste and many other amazing collaborators. Paper: Code:

GitHub - ermongroup/best-arm-delayed: Code for "Best arm identification in multi-armed bandits with...

Code for "Best arm identification in multi-armed bandits with delayed feedback", AISTATS 2018. - ermongroup/best-arm-delayed

github.com

1

7

19

Aditya Grover

@adityagrover_

6 months

Fully agree with @QuanquanGu . There is nothing limiting autoregressive (AR) models to plan prior to generating (action/word) tokens. In fact, any generative model can do it - AR, diffusion, etc. Our #NeurIPS2023 paper drives this message quite clearly

Decision Stacks: Flexible Reinforcement Learning via Modular...

Reinforcement learning presents an attractive paradigm to reason about several distinct aspects of sequential decision making, such as specifying complex goals, planning future observations and...

arxiv.org

Quanquan Gu

@QuanquanGu

6 months

I agree with Yann's view that current auto-regressive LLMs may not constitute the ultimate solution for achieving AGI. However, it's important to acknowledge that: Can LLMs do reasoning? Yes. LLM can implement first-order logic, see e.g., by Abulhair

17

50

411

0

3

17

Aditya Grover

@adityagrover_

6 years

For those of you attending #AAAI2018 , happy to chat more about two papers I'll be presenting: - Boosted Generative Models w. @ermonste (Sun, 2-3:30) - Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models w. Manik Dhar and @ermonste (Mon, 2-3:30)

0

3

17

Aditya Grover

@adityagrover_

2 years

@dustinvtran @ylecun +1 on probabilistic reasoning significantly advancing research in generative models. Autoregressive models, though, feel oblivious to much of PGM research from 2000s which focused on conditional independences and inference beyond density estimation (eg, marginalization).

0

16

Aditya Grover

@adityagrover_

1 year

Corrected blog link:

1

2

16

Aditya Grover

@adityagrover_

1 year

24 hours later: To be or not to be. "Real-time" editing of LLMs is hard!

Kate Crawford

@katecrawford

1 year

Umm, anyone a little concerned that Bard is saying its training dataset includes... Gmail? I'm assuming that's flat out wrong, otherwise Google is crossing some serious legal boundaries.

151

450

3K

2

0

16

Aditya Grover

@adityagrover_

10 months

Now with expanded support for new models, tasks, datasets, pretrained checkpoints, and more! Also check out our fresh preprint with benchmarking and analysis on robustness and transfer perf:

ClimateLearn: Benchmarking Machine Learning for Weather and...

Modeling weather and climate is an essential endeavor to understand the near- and long-term impacts of climate change, as well as inform technology and policymaking for adaptation and mitigation...

arxiv.org

Tung Nguyen

@tungnd_13

10 months

ClimateLearn, our PyTorch-based ML library for accessing climate datasets, state-of-the-art models, diverse evaluation metrics, and high-quality visualizations, just reached version 1.0.0! arXiv: Quickstart:

1

32

136

0

1

15

Aditya Grover

@adityagrover_

3 years

Check out the @berkeley_ai blog post on our recent preprint!

Kevin Lu

@_kevinlu

3 years

Complementary blog post to our paper, Pretrained Transformers as Universal Computation Engines, has been released!

0

18

58

0

15

Aditya Grover

@adityagrover_

6 years

A truly thoughtful and remarkable researcher+advisor! Congratulations @ermonste !

IJCAIconf

@IJCAIconf

6 years

A great pleasure to announce. Congratulations Stefano Ermon, @ermonste , Stanford University, the winner of the 2018 IJCAI Computers and Thought Award.

5

18

69

0

2

15

Aditya Grover

@adityagrover_

1 year

A key challenge with sequential decision making is that task specifications lack a universal API. Our latest work at #NeurIPS2022 looks at masking as a highly effective pretraining paradigm that enables flexible downstream RL for goal reaching, skill prompting, offline RL.

Hao Liu

@haoliuhl

1 year

In our #NeurIPS2022 work, we explore the generality of masked token prediction for generalizable and flexible reinforcement learning. A 🧵 on the paper

3

21

83

1

2

14

Aditya Grover

@adityagrover_

1 year

Congrats @tungnd_13 ! Very well deserved 🎉

Tung Nguyen

@tungnd_13

1 year

Very honored to be selected as one of the Amazon fellows at UCLA! And also thanks to my PhD advisor @adityagrover_ for his constant support. #AmazonScience #MINT

6

2

39

1

0

14

Aditya Grover

@adityagrover_

1 year

Developed with an amazing team of students: @tungnd_13 @hbXNov @_shashankgoel_ @jasonjewik @siddnandy Seongbin Park @tang_jingchen

1

2

14

Aditya Grover

@adityagrover_

2 years

@poolio Time to read only the titles

1

0

14

Aditya Grover

@adityagrover_

2 years

Check out our latest work that makes a first step using generative transformers for pretraining black-box optimizers: To appear as an Oral at the NeurIPS 2022 workshop on Foundation Models for Decision Making! w/ S. Krishnamoorthy, S. Mashkaria (2/2)

Generative Pretraining for Black-Box Optimization

Many problems in science and engineering involve optimizing an expensive black-box function over a high-dimensional space. For such black-box optimization (BBO) problems, we typically assume a...

arxiv.org

0

5

12

Aditya Grover

@adityagrover_

6 years

Submission deadline for the R2L Workshop in less than two weeks! We welcome novel research contributions, papers recently published or currently under review, position papers as well as papers introducing datasets and competitions! #NIPS2018

Aditya Grover

@adityagrover_

6 years

Website for our #NIPS2018 Workshop on Relational Representation Learning (R2L) is up! Workshop date: 8th Dec. Papers due on 19th Oct. @paroma_varma @fredsala #stevenholtzen @ProfJenNeville @ermonste @HazyResearch

1

34

67

0

3

12

Aditya Grover

@adityagrover_

6 years

Check out this blog post on our #AAAI2018 paper on Flow-GANs which takes a small step towards a principled quantitative evaluation of maximum likelihood and adversarial learning in generative models. Paper: Code:

GitHub - ermongroup/flow-gan: Code for "Flow-GAN: Combining Maximum Likelihood and Adversarial...

Code for "Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models", AAAI 2018. - ermongroup/flow-gan

github.com

Stefano Ermon

@StefanoErmon

6 years

New blog post on our #AAAI2018 FlowGAN paper by @adityagrover_ and Manik Dhar: surprising results comparing GAN vs. mixture of Gaussians! (btw Manik is applying to PhD programs - check out his folder if you are hiring)

0

28

70

2

12

Aditya Grover

@adityagrover_

3 years

@AlexGDimakis On its own, the discriminator is meaningless coz it is trained wrt a specific generator. But, in conjunction with the generator (post-training), we have found that it can parameterize an even more powerful generative model.

Bias Correction of Learned Generative Models using Likelihood-Free...

A learned generative model often produces biased statistics relative to the underlying data distribution. A standard technique to correct this bias is importance sampling, where samples from the...

arxiv.org

0

12

Aditya Grover

@adityagrover_

2 years

Generating abstract objects, such as task curricula, is challenging due to the plethora of qualitative desiderata and lack of any supervision. Our latest work shows that effective curricula can naturally emerge in appropriate multiagent scenarios. #ICLR2022

Yuqing Du

@d_yuqing

2 years

Humans excel at generating curricula of tasks simply by interacting with other agents. Can RL agents generate similar curricula? We tackle this in our #ICLR2022 paper, Multiagent Selfplay for Automatic Curriculum Generation! w/ @pabbeel , @adityagrover_ 1/8

1

34

172

0

1

12

Aditya Grover

@adityagrover_

6 months

Video-language alignment is critical for controlling agent behaviors in space and time. The large degrees of freedom for misalignment make naive data sourcing and scaling infeasible. What if we can generate and finetune specifically on misaligned scenarios? 🧵👇

Hritik Bansal

@hbXNov

6 months

📢 📽✍️We introduce VideoCon, a video-text dataset for training SOTA alignment model. It resolves a typical issue in video-text alignment models that struggles with robustness. w/ @YonatanBitton , Idan Szpektor, @kaiwei_chang , @adityagrover_ 🧵 1/

3

25

80

0

9

Aditya Grover

@adityagrover_

8 years

In spite of being average, Google Scholar was not really a choice until now. AI2 just made our lives easier!

0

4

9

Aditya Grover

@adityagrover_

5 years

Very cool work on continual learning for adaptive Internet video streaming! Also check out the paper:

Learning in situ: a randomized experiment in video streaming

We describe the results of a randomized controlled trial of video-streaming algorithms for bitrate selection and network prediction. Over the last eight months, we have streamed 14.2 years of...

arxiv.org

billions of packets

@justinesherry

5 years

For those of you who missed the news: my friend and colleague Keith Winstein at Stanford will let you watch US TV for free *legally* in your browser here: (And watching it helps networking research and science!)

5

61

205

1

3

8

Aditya Grover

@adityagrover_

6 years

So well deserved! Excellent example of rigorous empirical research. Congrats @alshedivat et al.!

OpenAI

@OpenAI

6 years

Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments (, tl;dr: agents that meta-learn outcompete agents that don't, ). Best paper award, Thursday May 3rd, 10:15-10:30am, Exhibition Hall A

3

18

61

0

2

8

Aditya Grover

@adityagrover_

4 years

@an_open_mind Evidence of the harmful biases of generative models in both language and vision is becoming an unfortunate new normal. Not all hope is lost: our recent work at #ICML2020 mitigates such biases for SOTA models *with NO label annotations*.

1

0

8

Aditya Grover

@adityagrover_

1 year

For representative work, see ClimaX: the first foundation model for weather and climate

Tung Nguyen

@tungnd_13

1 year

Introducing ClimaX, the first foundation model for weather and climate. A fast and accurate one-stop AI solution for a range of atmospheric science tasks. Paper: Blog: Thread🧵 #ML #Climate #Weather #FoundationModel

35

179

851

1

8

Aditya Grover

@adityagrover_

1 year

@SaggitariusA Corrected blog link:

0

1

8

Aditya Grover

@adityagrover_

3 years

Many thanks to our panelists @deviparikh , @james_y_zou , @radamihalcea , @sameer_ and @tatsu_hashimoto !

1

0

8

Aditya Grover

@adityagrover_

1 year

(2) ClimateLearn: open source PyTorch library for standardizing ML for climate science

Aditya Grover

@adityagrover_

1 year

📢Introducing ClimateLearn, a new PyTorch library for accessing climate datasets, state-of-the-art ML models, and high quality training and visualization pipelines. Blog: Docs: Quickstart Colab: 🧵 (1/n)

17

277

1K

1

7

Aditya Grover

@adityagrover_

3 years

@pranavrajpurkar @HarvardDBMI @harvardmed Congratulations, Pranav! Harvard is lucky to have you! :)

0

7

Aditya Grover

@adityagrover_

1 year

8 years ago, I developed one of the first (basic!) deep nets for wind forecasting w/ @akapoor_av8r @erichorvitz (KDD'15 -). With ClimaX, we can forecast/downscale/project any variable in arbitrary space/time better or comparable to operational systems!

A Deep Hybrid Model for Weather Forecasting | Proceedings of the 21th ACM SIGKDD International...

dl.acm.org

1

0

7