Abhijit Guha Roy

77 posts

Abhijit Guha Roy

Abhijit Guha Roy

@abzz4ssj

ML Engineer @GoogleHealth / Deep Learning Enthusiast / PhD at @AI_Medic

London, England Katılım Ağustos 2015
113 Takip Edilen208 Takipçiler
Abhijit Guha Roy retweetledi
David Stutz
David Stutz@davidstutz92·
Reliable ground truth is a key obstacle in developing AI systems. In a case study on medical diagnosis, we developed a framework for evaluation with ground truth uncertainty stemming from rater disagreement. This work was finally accepted at Medical Image Analysis – a thread 🧵:
English
1
9
34
4K
Abhijit Guha Roy retweetledi
Mayank Daswani
Mayank Daswani@mayankdaswani·
Cardiovascular disease is a leading cause of death globally, but access to screenings is uneven. Our research is attempting to change that. In a paper we recently published, we show how Photoplethysmography (PPG) signals can be used to predict long-term cardiovascular risk (1/5)
GIF
English
1
5
8
1.1K
Abhijit Guha Roy retweetledi
David Stutz
David Stutz@davidstutz92·
Delighted that our work on Monte Carlo conformal prediction has been accepted at @TmlrPub. This was a great collaboration of @GoogleDeepMind+@GoogleHealth to enable conformal prediction for problems where even experts disagree on the ground truth. 🧵 x.com/TmlrPub/status…
Accepted papers at TMLR@TmlrPub

Conformal prediction under ambiguous ground truth David Stutz, Abhijit Guha Roy, Tatiana Matejovicova et al.. Action editor: Fredrik Johansson. openreview.net/forum?id=CAd6V… #conformal #prediction #uncertainty

English
1
7
62
13.7K
Abhijit Guha Roy retweetledi
David Stutz
David Stutz@davidstutz92·
Here is our latest work on conformal calibration with ambiguous ground truth: We show how to perform conformal calibration in settings where it is difficult to obtain crisp ground truth labels from disagreeing annotations. More in the thread 🧵. arxiv.org/abs/2307.09302
English
1
7
59
7.1K
Abhijit Guha Roy retweetledi
Will Grannis
Will Grannis@GrannisWill·
What does #AI do when it doesn’t know? 🤔 @Google's @abzz4ssj & Ian Kivlichan have got the answers in last week's #Askatechspert, explaining how we teach algorithms to refrain from guessing and to ask for help from human experts in unfamiliar scenarios: buff.ly/3gwCfgG
Will Grannis tweet media
English
0
2
3
0
Abhijit Guha Roy retweetledi
Alan Karthikesalingam
Alan Karthikesalingam@alan_karthi·
🚨Wise doctors know when they don’t know - medical AI should too🚨New blog by @abzz4ssj @jessierenjie walks through our recent paper @GoogleHealth @GoogleAI @DeepMind enabling dermatology AI systems to detect the presence of conditions unseen in training goo.gle/31kUam7
Google AI@GoogleAI

Check out a novel approach to near-out-of-distribution detection with applications for dermatology image classification, which is better able to detect conditions previously unseen by the model during training. Learn all about it below ↓ goo.gle/3o3NOQA

English
0
22
89
0
Abhijit Guha Roy retweetledi
Pushmeet Kohli
Pushmeet Kohli@pushmeet·
The RVAI team @deepmind works on robustness and reliability of AI system and is hiring interns. If you are passionate about developing reliable and trustworthy AI systems (especially in NLP) and would like to work with us, please apply below by Dec 17th: deepmind.com/careers/intern…
English
1
8
18
0
Abhijit Guha Roy retweetledi
Alan Karthikesalingam
Alan Karthikesalingam@alan_karthi·
Our research @GoogleHealth @GoogleAI @DeepMind published at Medical Image Analysis goo.gle/31kUam7. Wise doctors know when they don’t know- medical AI should too. In dermatology this is critical, as many rare skin conditions occur too infrequently for AI to learn (1/n)
Alan Karthikesalingam tweet media
Google for Health@GoogleForHealth

In real-world deployment, medical #AI may encounter conditions it has never seen in training. Recognizing them is an important step to improving safety. Our new research demonstrates a novel method to tackle this challenge. goo.gle/31kUam7

English
5
59
288
0
Abhijit Guha Roy retweetledi
Google for Health
Google for Health@GoogleForHealth·
In real-world deployment, medical #AI may encounter conditions it has never seen in training. Recognizing them is an important step to improving safety. Our new research demonstrates a novel method to tackle this challenge. goo.gle/31kUam7
Google for Health tweet media
English
2
55
261
0
Abhijit Guha Roy retweetledi
Jie Jessie Ren
Jie Jessie Ren@jessierenjie·
We will be presenting "A simple fix to Mahalanobis distance for improving near-OOD detection" at UDL workshop @icmlconf tomorrow at 11:45am EST! 😀 arxiv.org/abs/2106.0902 Together with amazing co-authors @balajiln @stanislavfort @abzz4ssj @shreyaspadhy @jereliu
Jie Jessie Ren tweet media
Balaji Lakshminarayanan@balajiln

1/n Attending #ICML2021? Join us for our workshop on "Uncertainty and Robustness in Deep Learning" @icmlconf this Friday (July 23) 9am-5pm ET! ICML link: icml.cc/virtual/2021/w… Co-organized w/ @hendrycks @SharonYixuanLi @latentjasper @tdietterich @csilviavr @sebnowozin

English
3
10
50
0
Abhijit Guha Roy retweetledi
Florian Wenzel
Florian Wenzel@flwenz·
Don’t waste your models from hyperparameter search! Combining those models to form a "Hyperparameter Ensemble" leads to a performance much stronger than the single best model. We now released a notebook which shows step by step how to do it: bit.ly/3sp1m8G. 1/3
English
4
54
402
0
Abhijit Guha Roy
Abhijit Guha Roy@abzz4ssj·
Check out our new paper which tackles the challenging task of detecting previously unseen long tail of rare dermatological conditions encountered at test time. This is a crucial need for AI safety. A joint collaboration between @GoogleHealth @GoogleAI @DeepMind
Balaji Lakshminarayanan@balajiln

Excited to announce our new paper "Does Your Dermatology Classifier Know What It Doesn't Know?", led by @abzz4ssj @jessierenjie @jimwinkens along with an awesome set of collaborators @GoogleAI @GoogleHealth @DeepMind. Paper: arxiv.org/abs/2104.03829 Thread 1/n:

English
0
4
23
0

Keşfet

@TmlrPub @GoogleDeepMind @GoogleHealth @wonnga @NassirNavab @Google @jessierenjie @GoogleAI