Michael Chang @mmmbchang Twitter profile

Pinned Tweet

Michael Chang

1 month

It's such an honor to work on Project Astra with such an amazing team from across Gemini and Google DeepMind! While the #GoogleIO keynote was happening we had a last minute idea of watching the keynote with Project Astra. Check it out!

Google DeepMind

@GoogleDeepMind

1 month

We watched #GoogleIO with Project Astra. 👀

73

265

1K

15

23

354

Last Seen Profiles

@kGnZKhQhpGyTGWf

@tjw_97

@ShopeePH

@WiproCareers

@_caliluciano

@grame_rub

@KpOfg0x03g805ru

@LeszBuk

@CoolGundam

@patr1koo

@slam254

@Aira_srnk

@hijabjilbab1

@Russell02051998

@yukinyaan_darts

@FangCollective

@shanda_74

@REBELxoldier

@Roberto09234502

@FEST_HAL

@radosveta_vass

@FurryFiesta

@pscblh

@Tamakombu

@stw_pdg

@pamochan_home

@unbeugsamenfilm

@ccbrujil

@RIZWAN89436870

@sui_nayu

@JoePovall

@sun_vamps

@theelowkeey

@brivtny

@TamerAshour

@AquarianAstro

Michael Chang

@mmmbchang

1 month

Gemini and I also got a chance to watch the @OpenAI live announcement of gpt4o, using Project Astra! Congrats to the OpenAI team, super impressive work!

Michael Chang

@mmmbchang

1 month

It's such an honor to work on Project Astra with such an amazing team from across Gemini and Google DeepMind! While the #GoogleIO keynote was happening we had a last minute idea of watching the keynote with Project Astra. Check it out!

15

23

354

56

255

1K

Michael Chang

@mmmbchang

2 years

Learning to represent objects is a major research direction towards representing the causal structure of the world. In our oral at #iclr2022 workshop on Objects Structure & Causality, we present a new way to conceptualize objects: as stable points of a fixed-point procedure: 👇

5

48

327

Michael Chang

@mmmbchang

3 years

Does modularity improve transfer efficiency? In our ICML’21 paper (long oral), we analyze the causal structure of credit assignment from the lens of algorithmic information theory to tackle this question w/ @SKaushi16236143 , @svlevine , @cocosci_lab 1/N

2

37

170

Michael Chang

@mmmbchang

1 year

My dissertation talk can be viewed here: Thank you to my advisors @svlevine and @cocosci_lab for a PhD journey that has been the most intense and fulfilling growth experiences of my life. Thank you to all my friends and family for your love and support.

Neural Software Abstractions: Michael Chang Dissertation Talk

Title: Neural Software Abstractions: Learning Abstractions for Automatically Modeling and Manipulating SystemsSpeaker: Michael Chang (https://mbchang.github....

www.youtube.com

9

8

127

Michael Chang

@mmmbchang

9 months

Fun weekend project - a visualization of the generation process of a Bayesian flow network modeling a single point (-0.8, 0.8) (blue star). Pink dots are samples from sender dist, white traj are Bayesian updates over parameters mu, green traj are network's predictions over time.

NNAISENSE

@nnaisense

10 months

📣 BFNs: A new class of generative models that - brings together the strengths of Bayesian inference and deep learning - trains on continuous, discretized or discrete data with simple end-to-end loss - places no restrictions on the network architecture

6

117

511

4

15

127

Michael Chang

@mmmbchang

2 years

@jbhuang0604 Thanks for sharing! To add on, I've loved P. Winston's "How to Speak": , which shares a lot of elements you explained here.

0

23

126

Michael Chang

@mmmbchang

1 year

🤖🤖Multi-Agent Dialogue Simulations🤖🤖 Ever wondered how your favorite characters would interact in new contexts? How about putting Harry Potter and Argus Filch on the same team? New ex. in @LangChainAI for simulating stories of *multiple* characters

5

25

117

Michael Chang

@mmmbchang

6 months

♊️ Gemini is out! ♊️ What an honor it's been to work as part of the Gemini multimodal and eval teams with such amazingly talented and high velocity colleagues! Seeing its multimodal coding capabilities makes me ever more excited for the future of human-computer interfaces! 🚀

Demis Hassabis

@demishassabis

6 months

The Gemini era is here. Thrilled to launch Gemini 1.0, our most capable & general AI model. Built to be natively multimodal, it can understand many types of info. Efficient & flexible, it comes in 3 sizes each best-in-class & optimized for different uses

406

2K

12K

9

8

112

Michael Chang

@mmmbchang

1 year

🤖Data-Driven Character Chatbots🤖 Running out of creativity creating @character_ai character definitions? Introducing data-driven-characters, a repo built on @LangChainAI for easily creating, debugging, and running your own character chatbots grounded in any story corpus 🧵

3

16

107

Michael Chang

@mmmbchang

1 year

🔗 LangChain x Gymnasium 🤖 Chatbots have mostly been used as dialogue agents, but they can also be adapted for standard RL envs. New ex. in @LangChainAI showing how to integrate chat models with Gymnasium (formerly @OpenAI gym) from @FaramaFound (it doesn't do RL yet though).

5

21

102

Michael Chang

@mmmbchang

1 year

🗳️Decentralized speaker selection🗳️ How to implement multiagent dialogue without fixed schedule for who speaks when? Let the agents bid to speak! New ex. in @LangChainAI of a fictitious presidential debate b/w Donald Trump, Kanye West , Elizabeth Warren:

9

16

99

Michael Chang

@mmmbchang

4 months

Hard to contain my excitement about what we've been working on: it's exhilarating to see how quickly what would've blown my mind a short time ago -- like fine-grained analysis of hours of video -- become table stakes of what we expect from these models when things "just work."

Google DeepMind

@GoogleDeepMind

4 months

Introducing Gemini 1.5: our next-generation model with dramatically enhanced performance. It also achieves a breakthrough in long-context understanding. The first release is 1.5 Pro, capable of processing up to 1 million tokens of information. 🧵

47

426

2K

4

88

Michael Chang

@mmmbchang

4 years

1/ Check out our latest work on societal decision-making which we will present next week at ICML 2020. Very grateful to my collaborators and advisors @SKaushi16236143 , Matt Weinberg, Tom Griffiths, and @svlevine .

Sergey Levine

@svlevine

4 years

Can we view RL as a series of economic transactions between primitives/actions? We present an RL method based on auctions, where primitives "buy" states and "sell" next states w/ @mmmbchang , @SKaushi16236143 , Weinberg, Griffiths

2

72

391

3

11

74

Michael Chang

@mmmbchang

2 years

"Object Representations as Fixed Points: Training Iterative Refinement Algorithms with Implicit Differentiation" is now out on arXiv! w/ my advisors @cocosci_lab , @svlevine arxiv: web: youtube:

Michael Chang

@mmmbchang

2 years

Learning to represent objects is a major research direction towards representing the causal structure of the world. In our oral at #iclr2022 workshop on Objects Structure & Causality, we present a new way to conceptualize objects: as stable points of a fixed-point procedure: 👇

5

48

327

17

15

69

Michael Chang

@mmmbchang

1 year

🤖Dialogue Agents x Tools🔨 Equipping dialogue agents w/ browsing tools enables more grounded discussions. New ex. in @LangChainAI showing how to augment dialogue agents w/ tools in a fictitious debate between an AI accelerationlist & AI alarmist:

2

15

67

Michael Chang

@mmmbchang

1 year

🔗 LangChain x PettingZoo 🤖🤖🤖 Not only can you integrate LLMs with standard RL envs, you can now do so with multi-agent RL envs too. New ex. in @LangChainAI showing how to integrate chat models in PettingZoo (multi-agent Gymnasium) from @FaramaFound

Michael Chang

@mmmbchang

1 year

🔗 LangChain x Gymnasium 🤖 Chatbots have mostly been used as dialogue agents, but they can also be adapted for standard RL envs. New ex. in @LangChainAI showing how to integrate chat models with Gymnasium (formerly @OpenAI gym) from @FaramaFound (it doesn't do RL yet though).

5

21

102

0

17

57

Michael Chang

@mmmbchang

4 years

The code for Decentralized Reinforcement Learning: Global Decision-Making via Local Economic Transactions is now released: . Thank you to my collaborators and advisors @SKaushi16236143 , Matthew Weinberg, Tom Griffiths ( @cocosci_lab ), @svlevine !

GitHub - mbchang/decentralized-rl: Decentralized Reinforcment Learning: Global Decision-Making via...

Decentralized Reinforcment Learning: Global Decision-Making via Local Economic Transactions (ICML 2020) - mbchang/decentralized-rl

github.com

Sergey Levine

@svlevine

4 years

Can we view RL as a series of economic transactions between primitives/actions? We present an RL method based on auctions, where primitives "buy" states and "sell" next states w/ @mmmbchang , @SKaushi16236143 , Weinberg, Griffiths

2

72

391

0

10

54

Michael Chang

@mmmbchang

1 month

@arivero @OpenAI It wasn't a cut. Gemini abruptly stopped talking because I interrupted it.

3

0

48

Michael Chang

@mmmbchang

2 years

I'm now on the job market, looking for postdoc and industry positions. Interested in (1) developing AI for automatically modeling and manipulating systems (2) developing AI for powering next generation human-computer interfaces. If you know any opportunities, please let me know!

0

5

45

Michael Chang

@mmmbchang

2 years

How to repurpose previous knowledge for new problems is a major question in developing agents that automatically model and manipulate systems. In our oral at #NeurIPS2022 Attention workshop we study this question in the context of object rearrangement: 👇

Hierarchical Abstraction for Combinatorial Generalization in Object...

Contributed oral at the All Things Attention (https://attention-learning-workshop.github.io/) NeurIPS 2022 workshopAuthors:Michael [email protected]...

www.youtube.com

1

10

43

Michael Chang

@mmmbchang

7 days

Project Astra is featured in The Thinking Game, showcased at #Tribeca2024 film festival, featuring some members of the core Astra team!

Demis Hassabis

@demishassabis

8 days

Looking forward later today to the #Tribeca2024 premiere of The Thinking Game - a new documentary about the story of @GoogleDeepMind , AGI & AlphaFold, by Greg Kohs with music by Dan Deacon; it’s a sequel of sorts to the award-winning AlphaGo documentary

27

75

697

1

0

45

Michael Chang

@mmmbchang

1 year

🤖Two Agent Simulators🤖 Ever wanted to explore hypothetical variants of your favorite stories in role-playing games? Introducing a new example in @LangChainAI for simulating two agent dialogues, showing how to implement simple rpgs similar to D&D:

0

16

42

Michael Chang

@mmmbchang

1 month

Super excited to share what we've been working on at Google I/O!

Google

@Google

1 month

One more day until #GoogleIO ! We’re feeling 🤩. See you tomorrow for the latest news about AI, Search and more.

275

726

4K

0

2

41

Michael Chang

@mmmbchang

5 years

We will present our work with Abhishek Gupta, @svlevine , and Tom Griffiths @iclr2019 on Wed 11am #83 . Come see how composing representation transformations improves over learning flat input-output mappings when we want to extrapolate to harder compositionally structured problems.

0

8

40

Michael Chang

@mmmbchang

2 years

@TacoCohen What do you think is missing from how Konidaris et al. () define symbols? They define a symbol as the name for a set of percepts.

4

0

40

Michael Chang

@mmmbchang

5 years

Can agents learn to construct block towers from raw images by only learning about how blocks fall? See our spotlight and poster #ICML2019 with JD Co-Reyes @RndVar @michaeljanner @jiajunwu_cs @chelseabfinn Josh Tenenbaum @svlevine

2

6

33

Michael Chang

@mmmbchang

1 year

Missed the agent simulations panel at the summit last Fri? No worries, we got you. Here's an agent simulation of the agent simulations panel itself, letting you re-simulate the discussion w/ the panelists about any topic (w/ 🔊!)

Founders, Inc.

@fdotinc

1 year

stacked panel and a stacked crowd! agents and simulations panel is kicking off! @fablesimulation @JoinCultureDAO @DrJimFan @JackSoslow @mmmbchang @joon_s_pk @hthieblot

0

4

25

2

6

32

Michael Chang

@mmmbchang

5 years

Model-based RL in space of entities; entities processed w/ locally scoped functions & inferred from raw images through dynamic interaction without supervision on objects w/ @RndVar @JDCoReyes @michaeljanner @chelseabfinn @jiajunwu_cs J. Tenenbaum @svlevine

Sergey Levine

@svlevine

5 years

Model-based RL with models that factorize over entities; can discover object-like representations, and can be used to plan how to construct structures out of parts. w/ R. Veerapaneni, JD Co-Reyes, M. Chang, M. Janner, @chelseabfinn , J. Wu, J. Tenenbaum

0

23

105

0

10

31

Michael Chang

@mmmbchang

1 year

🖼️Solving visual analogies w/ in-context learning🖼️ In-context learning has mostly been shown in language; how can we transfer this capability to visual domain? Key challenge is learning tokens at appropriate abstraction. Our new paper led by @bhish_98 :

Im-Promptu: In-Context Composition from Image Prompts

Large language models are few-shot learners that can solve diverse tasks from a handful of demonstrations. This implicit understanding of tasks suggests that the attention mechanisms over word...

arxiv.org

Bhishma Dedhia

@bhish_98

1 year

Tired of engineering language prompts for your favorite text2img model? What if you could generate images directly from image prompts? Im-Promptu shows how you can learn to compose in-context from images; * no * language instructions required. Devil in the details 🧵 -->

1

7

24

1

7

30

Michael Chang

@mmmbchang

5 years

We have now released the code for our paper "Entity Abstraction in Visual Model-Based Reinforcement Learning" () here: .

Michael Chang

@mmmbchang

5 years

Model-based RL in space of entities; entities processed w/ locally scoped functions & inferred from raw images through dynamic interaction without supervision on objects w/ @RndVar @JDCoReyes @michaeljanner @chelseabfinn @jiajunwu_cs J. Tenenbaum @svlevine

0

10

31

0

2

29

Michael Chang

@mmmbchang

3 years

Check out our #ICML2021 long oral Thurs 7/22, where we apply causal analysis to the structure of RL algorithms to better understand transfer. w/ @SKaushi16236143 , @svlevine , @cocosci_lab Talk Poster Links (arxiv, youtube)👇

Michael Chang

@mmmbchang

3 years

Does modularity improve transfer efficiency? In our ICML’21 paper (long oral), we analyze the causal structure of credit assignment from the lens of algorithmic information theory to tackle this question w/ @SKaushi16236143 , @svlevine , @cocosci_lab 1/N

2

37

170

0

9

26

Michael Chang

@mmmbchang

4 years

Come check out our virtual poster session/Q&A for "Decentralized Reinforcement Learning: Global Decision-Making via Local Economic Transactions" tomorrow (July 16) at either 9:00am or 8:00pm Pacific Time

Sergey Levine

@svlevine

4 years

Can we view RL as a series of economic transactions between primitives/actions? We present an RL method based on auctions, where primitives "buy" states and "sell" next states w/ @mmmbchang , @SKaushi16236143 , Weinberg, Griffiths

2

72

391

0

4

22

Michael Chang

@mmmbchang

1 year

👑Authoritarian speaker selection👑 Instead of having all agents bid to speak, we can also have a privileged agent direct who to speak when. New example in @LangChainAI of how this can be done, with a fictitious Daily Show episode:

Michael Chang

@mmmbchang

1 year

🗳️Decentralized speaker selection🗳️ How to implement multiagent dialogue without fixed schedule for who speaks when? Let the agents bid to speak! New ex. in @LangChainAI of a fictitious presidential debate b/w Donald Trump, Kanye West , Elizabeth Warren:

9

16

99

2

6

22

Michael Chang

@mmmbchang

1 month

It has been amazing to work with @alexanderchen @suz_chambers and the rest of the team!

Alexander Chen

@alexanderchen

1 month

My teammate @mmmbchang had a lot of fun having Project Astra watch the #GoogleIO keynote live as it happened.🙂

5

0

18

0

1

22

Michael Chang

@mmmbchang

1 year

Here is a minimal re-implementation of Meta-Prompt by @noahdgoodman for building self-improving agents. It is written in @LangChainAI .

GitHub - mbchang/meta-prompt: A re-implementation of Meta-Prompt in LangChain for building self-i...

A re-implementation of Meta-Prompt in LangChain for building self-improving agents. - mbchang/meta-prompt

github.com

noahdgoodman

@noahdgoodman

1 year

Meta-prompt: A Simple Self-improving Language Agent

1

4

25

2

6

21

Michael Chang

@mmmbchang

3 months

✅ no waitlist ✅ 1M context ✅ free

Nikolay Savinov 🇺🇦

@SavinovNikolay

3 months

We removed the waitlist for Gemini 1.5 Pro with 1M context, try it for free in the UI:

17

34

265

1

21

Michael Chang

@mmmbchang

3 months

Both the process of knowledge creation and of programming rely on the cycle of conjecture and criticism, aka debugging. If AGI is capacity to create new knowledge, then the artificial programmer is the drosophila of AGI. Very excited for @cognition_labs on taking this first step!

Cognition

@cognition_labs

3 months

Today we're excited to introduce Devin, the first AI software engineer. Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork. Devin is

5K

11K

45K

0

1

20

Michael Chang

@mmmbchang

3 years

From Structure and Interpretation of Computer Programs:

0

5

20

Michael Chang

@mmmbchang

1 year

Here are some fantastic discussing the ideas in this @DavidDeutschOxf 's Beginning of Infinity: - @ToKTeacher 's ToKCast: - @VadenMasrani and @BennyChugg 's @IncrementsPod : - @ReachChristofer 's Do Explain:

Do Explain with Christofer Lövgren

Do Explain is a conversational podcast centered around questions of individual and collective human progress. Originally inspired by the visionary worldview of David Deutsch, the podcast has expanded...

doexplain.buzzsprout.com

Michael Chang

@mmmbchang

1 year

@MoritzW42 @sama @naval @nickcammarata @ToKTeacher @andy_matuschak @sharifshameem @moskov @maccaw @DavidDeutschOxf I read this the summer after my sophomore year based on @tejasdkulkarni 's recommendation and it changed my life. I credit this book with inspiring me fascination in research. I cannot recommend @DavidDeutschOxf 's book enough!

2

0

12

3

4

18

Michael Chang

@mmmbchang

3 years

From chapter 4 of @vmansinghka 's thesis:

0

1

16

Michael Chang

@mmmbchang

2 years

We will present our work next week at #NeurIPS2022 on "Object Representations as Fixed Points" on Nov 29 4:30-6:00 Central Time, Hall J Poster 505: w/ @svlevine and @cocosci_lab Paper, website, and talk below:

Michael Chang

@mmmbchang

2 years

"Object Representations as Fixed Points: Training Iterative Refinement Algorithms with Implicit Differentiation" is now out on arXiv! w/ my advisors @cocosci_lab , @svlevine arxiv: web: youtube:

17

15

69

0

2

16

Michael Chang

@mmmbchang

6 years

Sayan Gul was a wonderful friend to me. I always enjoyed my conversations with him, during which he made me feel how beautiful research was. Please consider donating to a new @cogsci_soc travel fund for undergraduates in honor of Sayan, who passed away on his way to #cogsci2018 .

Jess Hamrick is [email protected]

@jhamrick

6 years

If you attend @cogsci_soc , please consider donating to a new travel fund for undergraduates in honor of Sayan Gul, who was a student in my former lab and who tragically passed away this year on his way to #cogsci2018 .

0

16

0

3

16

Michael Chang

@mmmbchang

11 months

It was very fun to chat with @hwchase17 , @Metropolize_AI , and @AkashSamant4 at the @LangChainAI webinar and to see all the cool projects everyone has been working on! Some highlights that I've covered:

Harrison Chase

@hwchase17

11 months

There's huge demand for AI companions, as shown by @stuffyokodraws AI companion kit We recently hosted a webinar (w/ @Metropolize_AI , @AkashSamant4 , @mmmbchang ) on creating AI characters with LangChain - now on YouTube for some fun Friday viewing!

4

10

86

1

15

Michael Chang

@mmmbchang

1 year

Can ChatGPT generate ascii art? Let's find out! Some generations seem reasonable, but some are quite puzzling (further down the thread):

2

15

Michael Chang

@mmmbchang

1 year

Excited to give an invited talk @GRASPlab tomorrow about neural software abstractions! The talk will cover the following papers: - - - - and will feature my advisors & collaborators:

Decentralized Reinforcement Learning: Global Decision-Making via...

This paper seeks to establish a framework for directing a society of simple, specialized, self-interested agents to solve what traditionally are posed as monolithic single-agent sequential...

arxiv.org

GRASP Laboratory

@GRASPlab

1 year

Join us TOMORROW for a GRASP SFI presentation by Michael Chang from @UCBerkeley who will be presenting "Neural Software Abstractions: Learning Abstractions for Automatically Modeling and Manipulating Systems". For more info, please visit:

0

1

0

1

2

14

Michael Chang

@mmmbchang

6 years

Presented our poster with Sjoerd van Steenkiste on "Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions" at #ICLR2018 today.

0

14

Michael Chang

@mmmbchang

6 years

“Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions” (; ) at #ICLR2018 this Mon 11am-1pm, East Meeting level; 1,2,3 #13 . With Sjoerd van Steenkiste, Klaus Greff, Jürgen Schmidhuber.

1

2

13

Michael Chang

@mmmbchang

1 month

Credit to this "inception" idea goes to @jalayrac ! It's been a tremendous privilege working with you and the rest of the Gemini multimodal team!

JB Alayrac

@jalayrac

1 month

Congrats @mmmbchang ! So cool to see all the progress of Project Astra in the last months! Amazing to build these next gen Multimodal Models with such an awesome team!

4

3

31

0

12

Michael Chang

@mmmbchang

1 year

I am so lucky to have spent the past couple months with the amazing @LangChainAI team. I had opportunity to build on their incredible infrastructure to explore agent simulations () & data-driven characters. Thank you for the wonderful time, LangChain team!

Michael Chang

@mmmbchang

1 year

🤖Data-Driven Character Chatbots🤖 Running out of creativity creating @character_ai character definitions? Introducing data-driven-characters, a repo built on @LangChainAI for easily creating, debugging, and running your own character chatbots grounded in any story corpus 🧵

3

16

107

1

12

Michael Chang

@mmmbchang

2 years

6/ Yes we can: slot attention can be trained as a deep equilibrium model ()! We can use any root-finding solver to find the fixed point for the forward pass, and any method to directly compute the implicit gradient in the backward pass.

1

12

Michael Chang

@mmmbchang

1 year

@MoritzW42 @sama @naval @nickcammarata @ToKTeacher @andy_matuschak @sharifshameem @moskov @maccaw @DavidDeutschOxf I read this the summer after my sophomore year based on @tejasdkulkarni 's recommendation and it changed my life. I credit this book with inspiring me fascination in research. I cannot recommend @DavidDeutschOxf 's book enough!

2

0

12

Michael Chang

@mmmbchang

9 months

The code for reproducing this visualization, as well as for reproducing the other figures in Section 4 of the BFN paper (), can be found here: . Please let me know if you find any bugs!

GitHub - mbchang/minimal-bayesian-flow-networks: A minimal reproduction of the basic figures of the...

A minimal reproduction of the basic figures of the Bayesian Flow Networks paper for building intuition. - mbchang/minimal-bayesian-flow-networks

github.com

Michael Chang

@mmmbchang

9 months

Fun weekend project - a visualization of the generation process of a Bayesian flow network modeling a single point (-0.8, 0.8) (blue star). Pink dots are samples from sender dist, white traj are Bayesian updates over parameters mu, green traj are network's predictions over time.

4

15

127

0

3

11

Michael Chang

@mmmbchang

7 years

Presented our poster () at #iclr2017 today.

0

1

11

Michael Chang

@mmmbchang

4 years

This is one of the most inspirational papers I've read

Max Kleiman-Weiner

@maxhkw

4 years

New work from Kevin Ellis is a breakthrough for learning to synthesize programs. DreamCoder learns and extends a DSL and uses it to solve new problems faster. Rediscovers human-like languages: physical laws, vector algebra & functional programming:

0

54

278

0

1

11

Michael Chang

@mmmbchang

3 years

Come check out our work on "Modularity in RL via Algorithmic Independence" in #ICLR2021 ws! Generalization: 1pm PT: Learning to learn: 8:40am PT: . Grateful to my collaborators & advisors @SKaushi16236143 , @svlevine , @cocosci_lab

Sergey Levine

@svlevine

3 years

Can causality and algorithmic independence help RL transfer better? Tmrw, @mmmbchang will present "Modularity in RL via algorithmic independence" in #ICLR2021 ws: Generalization beyond... 1 pm PT: Learning to learn 8:40 am PT:

0

29

102

0

2

11

Michael Chang

@mmmbchang

7 months

I was amazed to see these demos internally and am super excited to see this work out in the world! Congratulations to the @GoogleDeepMind #Lyria team!

Demis Hassabis

@demishassabis

7 months

Thrilled to share #Lyria , the world's most sophisticated AI music generation system. From just a text prompt Lyria produces compelling music & vocals. Also: building new Music AI tools for artists to amplify creativity in partnership w/YT & music industry

111

536

3K

0

11

Michael Chang

@mmmbchang

5 years

Check out “Doing more with less: meta-reasoning and meta-learning in humans and machines” w/ Tom Griffiths, Frederick Callaway, @ermgrant , Paul Krueger, @FalkLieder where we argue that computation and data constraints are intrinsic to building intelligence

Falk Lieder

@FalkLieder

5 years

Our recent article “Doing more with less: meta-reasoning and meta-learning in humans and machines” is now available online:

0

2

18

1

0

10

Michael Chang

@mmmbchang

2 years

12/ There's more work to do to understand how implicit differentiation affects object centric models, but what is clear is that object-centric learning has potentially deep connections to other research areas (meta-learning, causality, fast-weights) that have yet to be explored.

1

0

10

Michael Chang

@mmmbchang

2 years

Great to see everyone at the poster session! Here were the posters presented today at the ICLR workshop on objects, structure, and causality.

Michael Chang

@mmmbchang

2 years

Learning to represent objects is a major research direction towards representing the causal structure of the world. In our oral at #iclr2022 workshop on Objects Structure & Causality, we present a new way to conceptualize objects: as stable points of a fixed-point procedure: 👇

5

48

327

0

2

10

Michael Chang

@mmmbchang

2 years

11/ One interesting finding is that the attention masks for implicit SLATE appear to be more smeared out. At first we couldn’t understand why, but folks at @genintelligent suggested that it could be learning to not only capture objects, but also their shadows.

1

0

10

Michael Chang

@mmmbchang

2 years

13/ Huge thanks to my advisors @cocosci_lab , @svlevine . And a shoutout to @shaojieb , @gautamsi11 for their help with deep equilibrium models & SLATE respectively. Talk Web 2 Papers ,

Object-Centric Learning as Nested Optimization

Our primary contribution is a proposal for a unifying tangible problem statement under which to view the set of iterative approaches to developed for object-centric learning so far.

openreview.net

1

0

10

Michael Chang

@mmmbchang

11 months

Looking forward to chatting with everyone at #ICML2023 !

0

9

Michael Chang

@mmmbchang

2 years

3/ Inferring from observation a set of representations that are a priori symmetric and independent requires a method for breaking symmetry. For many object-centric models, this is done through an iterative refinement process that is structurally similar to the EM algorithm.

1

9

Michael Chang

@mmmbchang

1 year

@LangChainAI Also check out @rahulchhabra07 's example () that also implements a similar bidding mechanism

Rahul Chhabra

@rahulchhabra07

1 year

Alice and Bob on a date. These are three independent AI agents trying to independently drive the conversation forward.

1

0

4

0

1

9

Michael Chang

@mmmbchang

2 years

5/ We empirically notice that slots of slot attention tend to remain generally stable. Can we treat these iterative algorithms as fixed-point procedures? If so, we can leverage recent advances in implicit differentiation techniques to stabilize training and improve learning.

1

9

Michael Chang

@mmmbchang

6 months

Excited to see Gemini out on @LangChainAI so quickly!

LangChain

@LangChainAI

6 months

🦜LangChain 🤝Gemini♊️ Gemini API access is out! Access it through LangChain with our first standalone integration package: `pip install langchain-google-genai` We're also launching an integration guide showing how to: 🎏Stream results 🖼️Use it's multimodal capabilities

17

162

680

0

1

7

Michael Chang

@mmmbchang

7 years

@jackclarkSF Godel, Escher, Bach by Douglas Hofstadter completely changed how I thought about the AI field

0

8

Michael Chang

@mmmbchang

3 years

Modularity is the capacity for components of a system to be independently modified. The most rigorous formalization of this we know comes from causality: modularity = algorithmic independence of mechanisms. (janzing & @bschoelkopf , 2010). 3/N

1

0

8

Michael Chang

@mmmbchang

4 years

@pcastr Thanks! The connection to Plato was inspired by this lecture by Patrick Winston:

Lecture 5.3: Patrick Winston - Story Understanding

MIT RES.9-003 Brains, Minds and Machines Summer Course, Summer 2015View the complete course: https://ocw.mit.edu/RES-9-003SU15Instructor: Patrick WinstonThe ...

www.youtube.com

0

2

8

Michael Chang

@mmmbchang

5 years

We are excited to have a range of interesting speakers at the #ICMLmodeling workshop () this Friday 6/14! Please submit your questions for our fantastic panel at the here: .

Generative Modeling and Model-Based Reasoning for Robotics and AI Panel Questions

Submit your questions for the panel of the workshop https://sites.google.com/view/mbrl-icml2019 here!

docs.google.com

0

3

8

Michael Chang

@mmmbchang

2 years

4/ The problem is that these iterative models have been difficult to train because they learn by backpropagating through an unrolled optimization procedure. The Jacobian norm of the slot attention cell (in red) blows up during training, leading to poor optimization.

1

0

8

Michael Chang

@mmmbchang

2 years

8/ One method is particularly simple: just truncate the backprop. This requires only one extra line of code and has O(1) space and time complexity in the backward pass. See also , , ,

1

0

8

Michael Chang

@mmmbchang

2 years

7/ It turns out that many different combinations of solvers and approximations for implementing implicit differentiation train much more stably than vanilla slot attention.

1

0

7

Michael Chang

@mmmbchang

6 years

Our work (w/ @TomerUllman , Torralba, and Tenenbaum) on the Neural Physics Engine is featured alongside Interaction Networks by @PeterWBattaglia , @DeepMindAI in @sciencemagazine as a step towards endowing intelligent agents with the same sense of intuitive physics as humans have:

Matthew Hutson

@SilverJacket

6 years

For common sense, #Ai must learn like a child. My feature in @sciencemagazine (Thanks @GaryMarcus , @ylecun , @DeepMindAI , @dileeplearning , @etzioni , @PeterWBattaglia , @frossi_t , @sapinker , @UNSW , @MIT , @allenai_org , Tenenbaum, Hinton, Thielscher, Goodman):

6

33

85

0

7

Michael Chang

@mmmbchang

6 months

For folks going to #NeurIPS2023 , @bhish_98 will present our latest work that shows how we can train models to make visual analogies through in-context visual prompting! Come check it out!

Bhishma Dedhia

@bhish_98

6 months

Excited to be presenting Im-Promptu at NeurIPS 2023! Catch us at Great Hall on Tue, 12/12, from 6:15 PM onwards. Will enjoy chatting about slot-centric methods and where this piece fits in the Generative AI puzzle 🧩

0

6

0

2

7

Michael Chang

@mmmbchang

2 years

2/ Objects reflect two very general properties about the causal structure of the physical world. The first is symmetry: the same physical laws apply to all objects. The second is independence: objects can be locally acted upon without affecting other objects.

1

7

Michael Chang

@mmmbchang

4 months

Multimodal error correction over code has been my dream for a while, and it's so wild to see @mckaywrigley demonstrate the beginnings of such capabilities with Gemini 1.5 Pro! Crazy to think that things like this will soon become the default that people expect from all models.

Mckay Wrigley

@mckaywrigley

4 months

The future of fixing bugs? Just record them. I filmed 3 separate bugs in an app and gave the videos to Gemini 1.5 Pro with my entire codebase. It correctly identified & fixed each one. AI is improving insanely fast.

219

1K

7K

1

0

7

Michael Chang

@mmmbchang

2 years

14/ Check out our posters at these #iclr workshops Apr 29: Objects, Structure, Causality: Gamification and Multiagent: Deep Generative Models for Highly Structured Data:

0

1

7

Michael Chang

@mmmbchang

3 years

Can we understand intrinsic motivation as an arms race between policies that minimize and maximize surprise? Check out this work led by @arnaudfickinger , @natashajaques , and @parajuli_samyak at the Unsupervised Reinforcement Learning workshop at #ICML2021 this Friday July 23!

Arnaud Fickinger

@arnaudfickinger

3 years

Effective unsupervised reinforcement learning requires a balance between seeking novelty and familiarity. How can we build an algorithm that strikes this balance? Paper: Project Page: Code:

3

19

78

1

0

7

Michael Chang

@mmmbchang

2 years

9/ Using implicit differentiation for slot attention improves qualitative reconstructions, as can we be seen from this comparison with the state-of-the-art SLATE architecture ()

1

0

7

Michael Chang

@mmmbchang

2 years

@stevesi "When you are thinking of the great ideas of history, often what you really want to be thinking about is the great representations that enabled people to think those ideas." -- Bret Victor ( @worrydream )

Bret Victor - The Humane Representation of Thought

For cool updated about the future of computing, follow me at https://twitter.com/iamvriad.Closing keynote at the UIST and SPLASH conferences, October 2014.Pr...

www.youtube.com

0

3

6

Michael Chang

@mmmbchang

3 years

Key idea: By representing the computation of learning algorithms as one giant algorithmic causal graph, we show that to get independently modifiable components, we need a credit assignment mechanism whose causal structure makes independent modification possible. 6/N

1

0

6

Michael Chang

@mmmbchang

1 year

Check out this very cool video generation platform from my good friends @ajayj_ and @_parasj !

Genmo

@genmoai

1 year

Announcing Genmo Video, a generative media platform with a new text-to-video model that can generate immersive live artwork from any prompt or any image. What will you create? 🎨▶️ Free public access: Discord: 👇1/n

15

63

252

0

6

Michael Chang

@mmmbchang

5 years

Paper and code for "Automatically Composing Representation Transformations as a Means for Generalization" available here:

Automatically Composing Representation Transformations as a Means for Generalization

Michael B. Chang, Abhishek Gupta, Sergey Levine, Tom Griffiths

sites.google.com

Michael Chang

@mmmbchang

5 years

We will present our work with Abhishek Gupta, @svlevine , and Tom Griffiths @iclr2019 on Wed 11am #83 . Come see how composing representation transformations improves over learning flat input-output mappings when we want to extrapolate to harder compositionally structured problems.

0

8

40

0

6

Michael Chang

@mmmbchang

1 year

Come see @a_li_d & @yayitsamyzhang present our work at @iclr_conf #ICLR2023 on object rearrangement on May 3: arxiv: youtube: w/ @a_li_d , @_kainoa_ , @cocosci_lab , @svlevine , @yayitsamyzhang Overview 🧵👇

Michael Chang

@mmmbchang

2 years

How to repurpose previous knowledge for new problems is a major question in developing agents that automatically model and manipulate systems. In our oral at #NeurIPS2022 Attention workshop we study this question in the context of object rearrangement: 👇

1

10

43

0

4

6

Michael Chang

@mmmbchang

5 years

Please consider submitting to our @icmlconf workshop on learning-as-model-building!

Aravind Rajeswaran

@aravindr93

5 years

Excited to announce our @icmlconf workshop on generative modeling and model-based RL! A step towards efficient real-world RL by drawing upon model-based methods. w/ Emo, Igor, Willie, Amy, Joelle, @mmmbchang , @doomie , @neuro_kim , @svlevine , Marvin.

0

20

65

0

2

6

Michael Chang

@mmmbchang

4 years

Please consider submitting and attending the @ORLR_Workshop at #NeurIPS2020 ! For more information, please check out the website . #ORLR2020

Object Representations for Learning and Reasoning

@ORLR_Workshop

4 years

Excited to announce the Object Representations for Learning and Reasoning at NeurIPS 2020! We will host child developmentalists, roboticists, and machine learning researchers on what objects are, what object representations should do, and what the challenges in applying them are.

1

15

54

0

1

6

Michael Chang

@mmmbchang

3 years

Paper: Website: Talk:

Modularity in Reinforcement Learning via Algorithmic Independence in...

Long Presentation at the Thirty-eighth International Conference on Machine Learning (ICML), 2021. Michael Chang*, Sidhant Kaushik*, Sergey Levine, Tom Griffi...

www.youtube.com

1

6

Michael Chang

@mmmbchang

3 years

“It is causality that gives us this modularity, and when we lose causality, we lose modularity.” -- @yudapearl , 4/N

1

0

6

Michael Chang

@mmmbchang

3 years

A popular hypothesis in machine learning is that modularity could enable efficient transfer. But it is an open question how to determine whether a learning algorithm is modular. Without a formal definition of modularity for learning systems, we can’t test this hypothesis. 2/N

1

0

6

Michael Chang

@mmmbchang

3 years

We can finally empirically test our hypothesis, and we find it survives the experimental test. Below, a modular on-policy rl algorithm (red) has higher transfer efficiency than its non-modular counterpart (blue). This trend appears consistent across many transfer topologies. 11/N

1

0

5

Michael Chang

@mmmbchang

1 year

@tobias_rees "Your first draft isn’t an unoriginal idea expressed clearly; it’s an original idea expressed poorly, and it is accompanied by your amorphous dissatisfaction, your awareness of the distance between what it says and what you want it to say."

1

0

5

Michael Chang

@mmmbchang

4 years

Very cool work by @michaeljanner , @IMordatch , @svlevine !

Sergey Levine

@svlevine

4 years

Gamma-models are dynamics models without a fixed time step. Instead, gamma models predict discounted averages of future state visitations, allowing us to train "infinite horizon" models with TD. w/ @michaeljanner & @IMordatch ->

6

20

192

0

5

Michael Chang

@mmmbchang

2 years

17/ Ask not what objects are, but what object representations enable an agent to do.

1

5

Michael Chang

@mmmbchang

4 years

8/ We show evidence that the local credit assignment mechanisms of our societal decision-making framework produce more efficient learning than the global credit assignment mechanisms of the monolithic framework.

0

5

Michael Chang

@mmmbchang

4 years

See our paper w/ @SKaushi16236143 , @svlevine , @cocosci_lab , NeurIPS DeepRL workshop Dec 11: causal analysis on structure of an RL algorithm, towards formalizing modular transfer in RL Poster: Paper: Video:

Michael Chang, Sidhant Kaushik, Sergey Levine, Thomas L. Griffiths · Modularity in Reinforcement...

slideslive.com

Sergey Levine

@svlevine

4 years

At deepRL WS, @mmmbchang will present “Modularity in Reinforcement Learning: An Algorithmic Causality Perspective on Credit Assignment” how causal models help us understand transfer in RL! Poster: Paper: Vid:

1

5

23

0

1

5

Michael Chang

@mmmbchang

2 years

10/ The quantitative metrics are also improved too. In terms of mean squared error, the implicit version has almost a 7x improvement over its vanilla counterpart.

1

0

5

Michael Chang

@mmmbchang

1 year

Neural Bucket Brigade largely inspired our work "Decentralized RL: Global Decision Making via Local Economic Transactions", which showed using the Vickery auction mechanism connects optimal local behavior to optimal emergent global behavior. Thread here:

Jürgen Schmidhuber

@SchmidhuberAI

1 year

Re: more biologically plausible "forward-only” deep learning. 1/3 of a century ago, my "neural economy” was local in space and time (backprop isn't). Competing neurons pay "weight substance” to neurons that activate them (Neural Bucket Brigade, 1989)

9

59

430

0

5

Michael Chang

@mmmbchang

1 year

It would be nice if we can generalize this customized character chat experience to allow the user to chat with any character based on any corpus, with full control over the design and data used to create the characters. Introducing data-driven-characters:

GitHub - mbchang/data-driven-characters: Generate chatbots from a corpus

Generate chatbots from a corpus. Contribute to mbchang/data-driven-characters development by creating an account on GitHub.

github.com

1

0

5