Besmira Nushi 💙💛
@besanushi
Followers
2,257
Following
757
Media
86
Statuses
1,569
Researcher @MSFTResearch artificial intelligence, responsible AI, technology & society.
Redmond, WA
Joined April 2010
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
BATALKAN BUKAN TUNDA
• 607333 Tweets
Polisi
• 594175 Tweets
FOTFOT NEW SINGLE
• 369614 Tweets
Mulyono
• 350145 Tweets
#SB19
• 343024 Tweets
#NHK_MUSICEXPO
• 334709 Tweets
Oprah
• 304399 Tweets
Gus Walz
• 193434 Tweets
El TSJ
• 143377 Tweets
Barron
• 52869 Tweets
Mikel Merino
• 47462 Tweets
Golpe de Estado
• 41016 Tweets
A'TIN ANG PANALO
• 40701 Tweets
Tribunal Supremo de Justicia
• 40013 Tweets
#BackToBackBillboardFanArmy
• 37484 Tweets
Eagles
• 37137 Tweets
Yale
• 32933 Tweets
#MAPATHEFIRSTTAKE
• 27792 Tweets
Pretender
• 25179 Tweets
Commanders
• 23402 Tweets
ちいかわ寿司
• 21804 Tweets
Dotson
• 18778 Tweets
#النصر_الرايد
• 18403 Tweets
Ann Coulter
• 12681 Tweets
チッケム
• 12454 Tweets
揖保乃糸
• 11632 Tweets
Milito
• 11508 Tweets
Pinned Tweet
New job post looking for senior ML Engineers in Model Evaluation and Understanding. If you are at
#ICLR2024
, come talk to our group at the Microsoft Booth tomorrow at 9:30 am. Link for application:
0
8
23
📣Happy to share that our group at
@MSFTResearch
is looking for Machine Learning researchers and engineers to advance the state of the art in understanding and evaluating foundation models (language and multimodal).
@ecekamar
10
50
203
The
@MSFTResearch
Summit is starting soon. We are happy to announce the release of the
#ResponsibleAIToolbox
as the very first open-source framework that combines RAI capabilities for assessing reliability, fairness, interpretability, and data quality in
#MachineLearning
. 1/n
3
28
141
Our internship job post on Evaluating & Understanding Foundation Models is out. Sharing here a list of open challenges our team is excited to explore together with future research interns. Application link:
@VibhavVineet
Neel Joshi
@hmd_palangi
@ecekamar
5
24
132
On Machine Learning errors and where to find them. Check out our tool on rigorous Error Analysis for
#ML
🛠️. We are excited to see this out in the open source community! Together with:
@ecekamar
@erichorvitz
@MehrnooshSam
@Azure
@MSFTResearch
@MxdRealityDev
3
33
119
The MSR Undergraduate Research Internship program is now accepting applications This is truly one of a kind opportunity for undergrads to get early experience and mentorship in research. Deadline November 19th.
4
39
106
Happy to share our job post for *in person* summer internships studying the robustness of platform models (Language & Vision). As our excitement on new large model grows, so does responsibility and need for in-depth understanding.
2
33
102
New research internship positions in the ASI group on studying and ensuring Robustness and Reliability for Vision and Language Models. Join us for an exciting research summer!
@hmd_palangi
@ecekamar
#MachineLearning
#NLP
#ComputerVision
0
30
89
Today we announce the release of two new open-source additions to Responsible AI Toolbox, for ideating mitigations in Responsible AI: *Responsible AI Mitigations library* and *Responsible AI Tracker* (1/n)
Behind tools for responsible AI is research for responsible AI. Learn how multidisciplinary teams come together across Microsoft to help practitioners on the ground develop AI that benefits people and society:
1
10
31
2
18
85
Resignation is not sufficient. The investigation needs to be repeated through a different set of experts. The previous committee needs to be questioned and investigated on how on earth they found the evidence insufficient. Resignation != Fired for sexual harassment
@UMich
0
4
61
Excited to share my invited talk at
#mlhc2020
this year: The talk summarizes ongoing work
@MSFTResearch
@Microsoft
on building tools for ML practitioners, rethinking engineering paradigms for ML, and designing ML in a human-centered way.
1
4
57
Our job post for internships on Failure Explanation for Machine Learning: . We encourage students working on trustworthy
#AI
and
#interpretability
for
#debuggingML
to apply and join our team during spring\summer 2020.
@ecekamar
@sytelus
@erichorvitz
1
13
47
The recordings of the Frontiers in Machine Learning event at
@MSFTResearch
are now available! Among others, we had a fruitful conversation with
@suchisaria
@tdietterich
@ecekamar
on ML reliability and robustness .
1
14
52
@timnitGebru
@JeffDean
Stay strong
@timnitGebru
, the whole AI ethics community is with you. This is a major step back, disrespectful for all the progress you're leading but also for the engineering workforce. Who protects the rights of engineers today, especially of those from underrepresented groups?
1
1
43
Thanks to all who made it to the
#NeurIPS2020
expo talk today! Among others, the presentation covered an intro to BackwardCompatibilityML . An OSS library & tool for 1) training model updates that minimize new errors, and for 2) visual model comparison. 1/n
2
7
47
1/n Our
#FAccT2022
paper shares insights on a real-world user study in the medical\radiology domain to better understand the impact of workflow design in medical decision making. An ongoing and super fun collaboration between
@MSFTResearch
@MarsGlobal
and Antech Imaging Services
1
5
45
I am giving a demo today at
#ICML2019
on our Error Terrain Analysis tool for test-time debugging of
#MachineLearning
models. Come talk to us at the
#Microsoft
Booth at 12:30-1:00.
@ecekamar
@erichorvitz
@MSFTResearch
1
6
42
On decomposing bias propagation in AI-advised decision making and the impact of equal representation as a bias mitigation strategy in
#Hiring
. Andi Peng talking about her
#AI
residency work in
@MSFTResearch
at
#HCOMP2019
.
@ecekamar
@koriinkpen
@ssuri
@emrek
@hcomp_conf
1
2
40
I'll be giving an update today at 2pm PT at NeurIPS Expo on our latest progress on tools for building reliable and responsible
#ML
. Q&A right after, hope to see you all there!
#NeurIPS2020
@MSFTResearch
@Azure
We'll be sharing a daily schedule of Microsoft presentations and live sessions at
#NeurIPS2020
. Attendees can visit our virtual booth to chat with experts and learn more about career opportunities.
Explore the full list of papers and upcoming events:
0
13
51
0
5
38
I will be at the Microsoft booth
#ICML2023
on Mon and Wed at 10 am to talk about Responsible AI Toolbox. Join us if you're interested. Also, please approach our booth if you are in the job market and you work in RAI problems (interpretability, reliability, safety, fairness)
1
7
38
Just about to start the tutorial on Guidelines for Human-AI Interaction at
#AAAI2020
@RealAAAI
Come meet us in Sutton South and take a pack of guidelines with you 😎
@adamfourney
Saleema Amershi
@dsweld
1
1
37
Excited to see the Responsible AI dashboard released as part of our larger efforts on the Responsible AI Toolbox for better integration of RAI tools. Take a peek at the research story behind these tools at Dev blog available at .
Responsible AI dashboard released! Excited to see the suite of tools come together. Stems from a great deal of thinking, building, refining tools aimed at assisting AI practitioners w/ fairness, intelligibility, reliability, and more
@besanushi
@MSameki
0
20
82
0
6
38
Last year we published a study in
@ICSEconf
on best practices & challenges for Software Engineering in Machine Learning based on a large survey in Microsoft. We have now shared the complete list of questions for the industry and the academic community here
1
12
35
Our latest research review on Frontiers of Multimodal Learning takes a responsible AI approach to model evaluation & improvement and makes 4 practical observations on insufficiencies of current methods and promising mitigation steps.
@MSFTResearch
blog at
New evaluation methods and a commitment to continual improvement are musts if we’re to build
multimodal AI systems that advance human goals. Learn about cutting-edge research into the
responsible development and use of multimodal AI at Microsoft:
1
4
15
1
7
36
Catching up with
#ICLR2020
ICLR keynotes.
@deviparikh
gave an exciting talk on learning from both vision & language, mutual grounding and modeling, biases, and better evaluation for human-AI teams. The VilBERT demo on common representations is impressive:
1
13
34
1/n The Responsible AI dashboard and scorecard are now available in
#AzureML
, leading the way for seamless integration of RAI capabilities with cloud analytics & computation for
#MachineLearning
. Read the amazing tech blog by
@m__s__paint
&
@MehrnooshSam
:
1
11
32
Happy to share our work on Backward Compatibility in Machine Learning at a dedicated tutorial on compatibility
@ICCV_2021
that summarizes recent work on the topic You can listen to this and other talks through the broadcast link:
2
5
34
Human-AI Interaction is not only a design concern. In fact, a lot more needs to be done in the
#ML
and
#SoftwareEngineering
stack to support effective interaction with AI. Check out our blog post on such considerations and what you can do to help today
1
7
31
It is devastating and outraging what is happening today in Iran...it is even more devastating to see how crimes against humanity are barely investigated, barely prosecuted at a global level in so many of these events.
#Mahsa_Aminii
1
6
33
New report on GPT4 evaluation from
@MSFTResearch
. What is most interesting from the paper is the very many creative approaches of designing novel tasks and tests. While these are not full benchmarks yet, collectively they constitute a new take on evaluation if applied at scale.
At
@MSFTResearch
we had early access to the marvelous
#GPT4
from
@OpenAI
for our work on
@bing
. We took this opportunity to document our experience. We're so excited to share our findings. In short: time to face it, the sparks of
#AGI
have been ignited.
67
728
3K
0
3
33
The Microsoft Research Summit starts on October 19 and is open to everyone. Among others, stay tuned for a full track on
#ResponsibleAI
during the summit and come join us for the lively talks, demos, and discussions.
Open to everyone! The first-ever 2021 Microsoft Research Summit, Oct 19 - 21, with over 150 sessions across 16 tracks, provides the global research community with an opportunity learn from experts pushing the frontiers of technology. Register now:
3
189
351
0
1
32
It was a pleasure to connect with the
#KDD
community at the
@ResponsibleAIK1
workshop and talk about Error Analysis, Barlow, Backward Compatibility as tools for ML debugging. Looking forward to more insights from further speakers in the schedule on related topics!
@MSFTResearch
Happening now!
Excited to learn about robust feature extraction for error analysis with
@besanushi
#raikdd2021
@kdd_news
0
0
7
1
8
31
1\N💡What do constraint satisfaction problems and factual queries to LLMs have in common? Our recent two works use a CSAT lens to evaluate and understand factual mistakes of LLMs in IR-style queries. It turns out that constrained queries are a major challenge even for sota LLMs.
1
10
30
Together with Mehrnoosh Sameki (
@MehrnooshSam
) we will talk about Error Analysis and the future vision around tools for
#ResponsibleAI
at
@odsc
#ODSCEast
. Come join us at 12:40 PDT for the talk or for more questions at the
@Azure
booth.
ODSC East Virtual Conference Speakers:
#ODSCEast
#DataScience
#MachineLearning
@besanushi
@MehrnooshSam
@Microsoft
0
7
9
0
7
29
What are the properties of a good AI collaborator? Gagan
@bansalg_
is giving a talk on how the complexity and the stochasticity of the machine learning error boundary affect Human-AI Collaboration.
@hcomp_conf
#HCOMP2019
#AI
The paper is up here:
1
8
27
Read Megha's story on the MSR AI residency program. Megha is a bright soon-to-be PhD student at Stanford and has decided to spend a year at
@MSFTResearch
working on challenging problems in machine learning such as backward compatibility in ML and model comparison.
0
3
28
Amazing news today
@MSFTResearch
:
@susan_dumais
has received the SIGCHI Lifetime Research Award as well as
@merrierm
and Steven Drucker have become inductees of the CHI Academy. Way to go!
#FridayMorning
#Celebration
#HCI
0
1
27
Great to see this World Economic Forum highlighting FairLearn, InterpretML, ErrorAnalysis to show the impact of tools on RAI. "Responsible Use of Tech: The Microsoft Case Study"
@MehrnooshSam
@ecekamar
@hannawallach
@MiroDudik
@HarshaNori
@erichorvitz
1
4
25
Giving a talk/demo today on Error Terrain Analysis for ML, a collaboration with the Ethics& Society + Cognition teams. Come see us at
#DebugML
@iclr2019
R03 (10:40). Towards automated tools for robust ML engineering.
@MSFTResearch
@ecekamar
@erichorvitz
1
1
24
Hey everyone, at MSR AI we're looking for a bright intern to work with us on Failure Explanation and Debugging of machine learning systems. To learn more about the amazing internship experience at
@MSFTResearch
pass by Booth 203 at
#NeurIPS2018
1
10
22
New open positions in our group on building human-centered experiences and tools for
#ResponsibleAI
. You'll be working with a fantastic set of people and be able to make an impact on real-world AI products.
We're hiring! Multiple opportunities to join the Human-Centered AI group
@MSFTResearch
to create
#ResponsibleAI
experiences and tooling. Apply now at the links below for Researcher, Postdoc, and Research Software Engineer positions.
#HCI
#AI
#HumanCenteredAI
#Microsoft
2
143
430
0
7
25
How can we discover generalizable failure modes in deep learning models for image data? Barlow is an error analysis framework that extracts interpretable visual features from images and uses them to explain model failures to
#ML
practitioners. Soon in
#CVPR2021
Introducing Barlow, a framework for finding bugs of deep neural networks.
Paper:
Code:
Joint work with
@besanushi
@ecekamar
@sytelus
@erichorvitz
Some examples of bugs discovered by Barlow in the thread ⬇️.
1
28
105
1
4
24
Excited to see this work being presented at
#AAAI19
. Together with
@bansalg_
@ecekamar
@dsweld
@wslasecki
and
@erichorvitz
, last summer we put a lot of thought into improving model updates so that they do not violate human trust or even contracts with other software components.
At
#AAAI19
:
@MSFTResearch
&
@UWcse_ai
team identify an important challenge in human-AI collaboration: Updates to
#AI
components (even good ones) not understood by end users. Pursuing “backward compatibility" of AI
@realAAAI
@acm_sigai
@sigchi
@usermodeling
0
6
14
0
0
22
Join us tomorrow for the TrustML social. The agenda contains games, panel discussions, and mentoring sessions. 3pm - 5pm ET / 12pm - 2pm PT / 9pm - 11pm UTC+1
#NeurIPS2020
#TrustworthyML
Reminder to
@NeurIPSConf
attendees: our Social is scheduled just about 24 hours from now: 3-5pm ET tomorrow! Tune in for some exciting virtual activities and discussion. Joining info in the main thread below👇🏽
0
8
37
0
4
24
@FrancescoLocat8
's keynote at the Spurious Correlations workshop
#ICML2023
emphasizes that OOD is still a major issue even in large foundation models. This is such an important observation in a time when the general belief is that internet-scale data will "solve" robustness.
1
3
25
Some of the best computer scientists I have met and are friends with have studied at
#SharifUniversity
. Today, this university is under attack by a ruthless dictatorial regime that is targeting the brightest of people. Be their voice, social awareness matters
#IranProtests
0
5
21
We're giving a tutorial today at
@FAccTConference
on Responsible AI Toolbox together with Rahee Ghosh Peshawaria and Wenxin Wei. Come join us for demos, and a conversation on challenges of building and adopting tools for Responsible AI.
1
5
23
Check our latest work on how to prioritize the accuracy optimization of
#ML
models when it matters, aka when it is more likely for
#people
to trust the model. Together with:
@bansalg_
@ecekamar
@erichorvitz
@dsweld
@MSFTResearch
#AAAI2021
1
4
24
#MSFTFacSumm
has started! Dan Weld is talking about machine learning debugging and intelligibility, including our recent work with
@bansalg_
on backward compatibility in
#ML
for human-centered model updates.
0
2
23
Once again, the suspension of H1b visas feels like an excuse for derailing the conversation from "what can we do to mitigate the pandemic" to "whom can we blame this time".
Now is not the time to cut our nation off from the world’s talent or create uncertainty and anxiety. Immigrants play a vital role at our company and support our country’s critical infrastructure. They are contributing to this country at a time when we need them most.
183
726
3K
2
3
20
Happy to see the MS work on
#ResponsibleAI
being released publicly as a resource center! Resources include practical approaches for timely problems: Fairness, ML Debugging and Intelligibility, Human-AI Collaboration, Privacy, and just good old/new engineering practices for AI.
Microsoft rolls out a Responsible AI Resource Center:
@PartnershipAI
@RealAAAI
@WHOSTP
@AiCommission
@theNAEng
@IEEEorg
#responsibleAI
#MicrosoftAether
6
46
142
0
0
21
Thanks to
@gr3tchen
for hosting me in the
#MSRPodcast
! It was a lot of fun talking about the work we do at
@MSFTResearch
on Human-Centered and Trustworthy AI. Of course, so much left to be done 🛣️! Side note: the podcast also has many other great episodes, definitely recommended!
Dr. Besmira Nushi envisions AI as a collaborator with humans—all of us. Her work in the Adaptive Systems and Interaction group aims to improve processes that make AI reliable & safe for everyone. Hear how
@besanushi
navigates this challenge on
#MSRPodcast
:
2
42
229
1
2
21
It is not the first time this happens but I do not understand why so many academic deadlines are either during the weekend or on a holiday. The camera ready deadline for
@TheWebConf
is on *Sunday* (Jan 26th). How would it change the publishing process if it was a day later?
4
0
21
As
#AI
is being deployed in products, we advise that it's important to rethink design practices to accommodate fluid and trust-based human-AI interaction. Each of the guidelines is an emergent research direction, hope to see more work on this in the
#ML
\
#HCI
communities.
#CHI2019
New Guidelines for Human-AI Interaction mark an important step towards much-needed best practices for designing helpful, responsible, and trustworthy AI:
#CHI2019
0
77
462
0
4
21
Data contamination is being discussed extensively, especially in the age of closed science in ML. To a certain extent, contamination is now intentional because 1) it is hard to prove, 2) it still manages to impress in a less scientific but profitable way.
1
3
22
Saleema presenting "Guidelines for Human-AI Interaction"
@chi2019
. A great resource for both AI and UX practitioners as they're designing the next AI technologies.
#CHI2019
@MSFTResearch
Read more at:
0
2
21
Registration is open to all at . The
#ResponsibleAI
track lead by
@hannawallach
&
@ecekamar
covers recent work and thoughts on the topic, including a demo on the
#ResponsibleAIToolbox
. Airing at 11am CST, 10am BST, 9am PT. Join us for the demo and Q&A! 2/n
1
7
20
Join us for our webinar on bringing RAI principles to practice with rigorous RAI tools. Registration link at: Together with
@ecekamar
, We will be talking about Error Analysis () and BackwardCompatibilityML ()..
Making AI safer and more reliable is essential for its success.
@besanushi
&
@ecekamar
will discuss principles for responsible AI Mar 18 at 10 AM PT. Explore development tools including error analysis, model comparison and backward compatibility. Register:
2
22
83
0
7
20
Our latest
#ICML2023
paper provides a flexible & computationally efficient solution to address spurious correlations of vision-language models, through language specifications. Work lead by
@YUYANG_UCLA
during her
@MSFTResearch
internship, together with
@hmd_palangi
@baharanm
🎙️We are thrilled to announce that we will be presenting our latest paper with
@besanushi
@hmd_palangi
@baharanm
() at
#ICML2023
! 🎉 Join us as we share insights and solutions for ✨spurious correlations in vision-language models✨. (🧵1/8)
10
5
94
0
1
21
Looking forward to learn more about the new projects that
@erichorvitz
will lead as Microsoft's Chief Scientific Officer.
@MSFTResearch
thrived in the last years in founding&advancing important areas as Responsible AI, ML Theory, Human-AI Collaboration, Integrative AI
#BigThanks
!
Some news on taking on a new responsibility
@Microsoft
Much work to do. Interesting challenges & exciting opportunities ahead
@MSFTResearch
@theNASEM
@NSF
@NIH
@DARPA
@nlm_news
@PartnershipAI
@AiCommission
@kevin_scott
@satyanadella
@allen_ai
@StanfordHAI
38
39
378
0
0
19
A must read on fascism and soon to be available in English (out on August 4th). Satirically and concisely written by Michela Murgia
@KelleddaMurgia
How to Be a Fascist: A Manual
#BlackLivesMatter
#Fascista
ÈChiIlFascistaFa
#KnowYourEnemy
1
3
19
We're welcoming
@icmlconf
attendees to the Human-AI Collaboration for Sequential Decision Making workshop together with Sebastian Tschiatschek and Adish Singla (
@adishs
). Speakers will cover a variety of topics at the intersection of
#HumanAICollaboration
#ReinforcementLearning
1
1
20
Andi Peng presenting at
#TRAIT2022
#CHI2022
on studying the dynamics of performance & bias in joint human-AI decisions for
#hiring
. More accurate models always boost human performance, BUT human reliance on AI is model-specific which results in very different
#bias
propagation.
Excited to present an encore of our AAAI oral at
#CHI2022
#TRAIT2022
today!
As we continue to deploy real AI recommender systems in the world, it is important to understand their impact on human BIAS and ACCURACY.
2
7
32
0
2
20
Exciting opportunities in
#fairness
for
#MachineLearning
and large-scale language models. Roles expand across different levels and locations
@MSFTResearch
.
🚨📢 Super exciting news!!!! 🎉🎊 I'm hiring applied scientists to join my team to kickstart a new initiative on the fairness of large-scale language technologies!!!! For more info or to apply, see:
9
138
454
0
3
19
So fun to be together with so many great women in AI at
@Microsoft
!
@hannawallach
@jennwvaughan
@katjahofmann
@besanushi
0
1
23
0
1
18
This year at
#NeurIPS2019
: Project Petridish: Efficient forward neural architecture search on learning how to efficiently grow and extend a parent model to a model with higher capacity.
@Eryuelan
@debadeepta
@erichorvitz
@MSFTResearch
0
1
18
Human-Centered AI Design by
@mihaela_v
By the way, the article also contains practical printouts for flashcards and posters. Take a look! We're eager to listen to your thoughts.
@MSFTResearch
0
6
19
War day seven: 9100 people would have still been alive today if it wasn't for the insanity of a whole warfare and propaganda machine.
#StandWithUkraine
0
1
18
When it comes to politics, method and contents coincide, and the fascist method holds the power of alchemical transmutation: if applied without any ideological prejudice, it turns whoever makes use of it into a fascist, because—as Forrest Gump would say—fascist is as fascist does
1
0
18
This whole series of events where one buys a weapon in a shop, shoots many people, and official authorities call it a bad day or a non-racial act, is a sad testimony to how questionable policies and harmful political propaganda still deeply dominate in the country.
#atlantaspa
0
1
17
New vision application enables
#InclusiveTravel
through
#AI
for helping railway passengers with impaired vision. The team analyzed models from a
#ResponsibleAI
lens, using
#ErrorAnalysis
for reliability assessment and
#CoreML
for private inference.
@Azure
0
3
18
This year at
@kdd2020
: "An Empirical Analysis of Backward Compatibility in Machine Learning Systems". Work lead by
@megha_byte
deep dives into understanding why backward incompatibility happens in Machine Learning and how updates in
#ML
models can negatively impact larger systems
Beyond accuracy: Study
@kdd2020
on how updates aimed at improving
#ML
models may lead to new errors with downstream influences. Findings impt. for maintaining systems
@besanushi
@ecekamar
@MSFTResearch
@AMIAinformatics
@PartnershipAI
@AiCommission
#AI
0
6
30
1
3
18
Learn how to build responsible Human-AI experiences with
#haxtoolkit
.
@mihaela_v
and
@SaleemaAmershi
will cover all aspects of the toolkit on July 21st at 10 am PT.
@MSFTResearch
#HumanAI
We* made a thing and can't wait to share it with you! Join us for this webinar!
#haxtoolkit
*
@SaleemaAmershi
@adamfourney
@hannawallach
@jennwvaughan
@besanushi
@megyoung0
1
7
34
0
1
18
While photorealism has certainly improved in DALL-E 3, seems like little has changed in terms of representational fairness. Had hoped that the figures we presented earlier at
@AIESConf
this year on DALL-E 2 and Stable Diffusion would improve ().
1
4
18
LIDA is now available as open source. Try it out with your own data today!
🎉 Thrilled to announce the OSS release of LIDA - a lib for automated visualization with
#llms
🛠️ API for
#datavisualization
generation, eval, editing, recommendation
🔄 Supports multiple LLM providers (
@openai
,
@cohere
, PaLM,
@huggingface
)
🌟 Great UI
➡️
12
54
205
1
3
18
@RealAAAI
@adamfourney
@dsweld
All slides available here: Wish we could also document all the good questions that came up during the Q & A thanks to the amazing audience!
2
8
16
Our
@AIESConf
2023 paper quantifies "Social Biases Through the Text-to-Image Generation Lens" across gender, race, age, and geographical locations. TLDR of findings below 1/N
1
7
18
@mmitchell_ai
@dinabass
@nitashatiku
@Abebab
@laion_ai
@SashaMTL
@ria_kalluri
@Leonardonclt
It is so important to point out these concerns as we move forward in the generative space. Generative settings amplify biases in ways that are much worse than reality or retrieval methods. We found out similar results earlier this year for our AIES work:
1
6
16
Join us today at 10:00 AM PT, for a deep dive into the Responsible AI Dashboard during the
#ODSC
conference. We will go through hands-on demos for debugging machine learning models and causal decision-making.
@MehrnooshSam
#ResponsibleAI
#MachineLearning
Join FREE ODSC Webinar to learn how to accelerate engineering processes in machine learning by enabling practitioners to design customizable workflows and tailor RAI dashboards that best fit with their scenarios. Register now:
#ODSCWebinar
@Microsoft
0
0
2
0
3
14
The third day of the
#MSFTFacSumm
is featuring a diverse set of workshops on the Future of Work.
@vanimt
and I are hosting a workshop on research directions for large-scale citizen science and human-AI collaboration.
0
4
16
@thegautamkamath
Actually...THAL-E would be a much better AI jf it would convert descriptions to real Indian food 😇
#greedy
#agi
#aiforfood
1
0
15
I'd like to see how this AGI will find my missing socks and what reward function it is going to use to do that. Is it going to mess up all closets to look for a sock? Or is it going to apply my lazy greedy search? Also, wondering if we're abusing the term
#AGI
too much these days
DeepMind says reinforcement learning is ‘enough’ to reach general AI
0
2
22
0
0
16
I have seen a lot of people doubting the problematics of lack of diversity in AI-generated content. This article uses a generated photo, illustrating humans telling stories through data (quite compelling) but also subtly conveys that only men can do so... 1/N
2
2
15
To all
#ML
practitioners: What are the top-3 tools and libraries that you use today to make sure that you are building
#ResponsibleAI
? Examples may include tools for model assessment,
#debugging
,
#fairness
,
#interpretability
,
#robustness
. 2-min survey:
0
9
13
See 8 challenges of evaluating text generation from Asli Celikyilmaz
@real_asli
. These reflect how complex evaluation can be in such systems but also how misleading current evaluation practices and metrics can be if they do not consider similar issues in depth.
That was it from the keynote from
@real_asli
at the GEM workshop, including this very relevant slide.
We are taking a 5 minute break and are then back with our panel.
0
3
18
0
0
13
The redesigned Bing Translator tackles gender bias by offering more than one translation option. The team also released GATE: a linguistically diverse corpus of gender-ambiguous source sentences with multiple alternative target language translations:
1
1
14
@ICSEconf
The full paper can be found here: together with: Saleema Amershi
@abegel
@christianbird
Rob DeLine, Harald Gall
@ecekamar
Nachi Nagappan
@tomzimmermann
@MSFTResearch
0
2
13
@aselipsky
Open source code and tool was also released via the BackwardCompatibilityML package: containing a series of loss functions in
#TensorFlow
and
#PyTorch
as well as an interactive widget in
#jupyter
for model exploration and comparison.
@ecekamar
@erichorvitz
0
6
14
Check it out: 🥳and contact us if you have new examples to add to the HAX Design Library.
#HumanCenteredAI
It's here! Today we launched the
#HAXToolkit
, practical tools for creating responsible Human-AI eXperiences.
This was a labor of love, and sometimes sweat and tears, by many
@Microsoft
. This is the
#HAX
story. Opinions my own.
#AI
#ux
#responsibleAI
2
71
188
0
2
14
It is the 18th century all over again, perhaps every day, even when we don't see video footage of others murdered like
#GeorgeFloyd
. Racism is not over. Never been over.
#BlackLivesMater
0
2
13
April 27 and 28: Workshop on Platform for Situated Intelligence
#psi
. In-depth tutorials, debugging visualizations, and success stories. If you're building complex
#ML
systems beyond single models, you'll feel yourself home.
Register to attend our Platform for Situated Intelligence Workshop on April 27–28. You'll see tutorials and in-depth discussions on an open-source framework that accelerates development and research on multimodal, integrative-AI applications:
0
21
95
0
1
13
I had a joke about explainable ML but I had to explain it.
0
2
14
#3facts5scientists
1. Rented a computer to complete my first CS course project.
2. Didn't know there's a diversity issue in CS until age of 25 😱.
3. Wanted (still want) to be a librarian as a kid.
@JanaGiceva
@anjagruenheid
@debadeepta
@krrun16
@ecekamar
1
0
13
The toolbox integrates ideas & technology from mature RAI tools:
#ErrorAnalysis
#Fairlearn
#InterpretML
#EconML
#DiCe
lead by several groups across
@MSFTResearch
&
@Azure
. It enables ML practitioners to combine assessment capabilities into customizable dashboards. 3/n
1
2
13
The toolbox could not have been possible without the passionate work on research, design and engineering of RAI leaders:
@MehrnooshSam
@romanlutz13
@MiroDudik
@m__s__paint
Ilya Matiach
@ecekamar
@hannawallach
@erichorvitz
@HarshaNori
Rich Caruana
@amt_shrma
@syrgkanis
ALICE 4/n
2
0
13