Gautam Shroff
@gmshroff
Followers
482
Following
137
Media
1
Statuses
76
Further to @GaryMarcus and other who rightly point out that #DeepSeek is 'just' more efficient, not 'better', one wonders if having lots of cheap agents debating can give 'better' results: enough for researchers to do now, even without the big $$!
Nice article in the Indian Express where I have also shared some thoughts. The larger message to the research community especially in India is perhaps that we should not assume that the 'big labs' cannot be bested with brains from wherever!
0
0
0
Nice article in the Indian Express where I have also shared some thoughts. The larger message to the research community especially in India is perhaps that we should not assume that the 'big labs' cannot be bested with brains from wherever!
🧷We asked AI researchers in India what DeepSeek’s disruption means for us. Even as the US and China lead the AI race, they say we can catch up. “India could do a Mangalyaan in AI,” said @gmshroff, professor at IIIT-Delhi who earlier led research at TCS. A thread 1/4
0
0
2
Turned 60 and 'retired' from @TCSResearch. But not 'tired': joined @IIITDelhi as a professor. Some research to continue - analogy (e.g. https://t.co/OG5gmewEYS), finance (e.g. https://t.co/BXrB65ulEl). Plus working on an ed-tech initiative, and hope to post more often too.
7
3
81
I enjoyed this conversation with Kunal from @AnalyticsVidhya with his well informed questions.
Dr. @gmshroff, TCS Fellow, SVP & Head, TCS Research, @TCS, shares his POV on the potential of #AI, insights about a career in #research & more, in conversation with @jainkunal, @AnalyticsVidhya. Watch the episode here - https://t.co/v4qGl5bR6H
#PhDChat
0
0
2
Kudos to my collaborators especially students and faculty at BITS Goa: three papers at the upcoming IJCLR conference including this one. We are looking for pre-doc/post-doc applicants in Neurosymbolic reasoning. @TCSResearch @SforAiDL
Knowledge-based Analogical Reasoning in Neuro-symbolic Latent Spaces https://t.co/uzufNcs8Ma by Vishwa Shah et al. including @gmshroff
#NeuralNetwork #DistributedRepresentations
1
6
28
Also, are we trying to 'understand' brains or solve practical problems? For the latter, symbols, programs or networks - only performance matters. For the former, there is a certain appeal of the purely neural approach. Distinguishing goals is important IMO.
0
0
5
Excellent discussion of the symbol debate: "search' vs. an end-to-end NN trained via GD. Question: could 'search' (over whatever, compositions of neural modules or symbolic programs) be executed via a recurrent network with indeterminate number of steps? @GaryMarcus @ylecun
My thoughts on the symbolic/non-symbolic AI debate @GaryMarcus @ylecun, which I think is maybe not the right question to address:
2
2
4
Continual learning is usually motivated by trying to mimic plasticity/stability in brains. Here we present a practical motivation in an industry-inspired setting where task dimensions also vary. #inventingforimpact @TCSResearch @ylecun
Continual Learning for Multivariate Time Series Tasks with Variable Input Dimensions https://t.co/CWu1L2XTOz by Vibhor Gupta et al. including @gmshroff
#Classifier #RecurrentNeuralNetwork
0
0
2
40 years ago, this month, TCS Research was born. We salute the visionary who started the Software Research Lab in @TCS - Prof. Kesav Vithal Nori. Click to hear Professor Nori speak about how it all began - https://t.co/0AX9W0S9Cq
#TCSResearch #Research #TCS
#40andforward
0
7
40
Nice work! Consistent gains on many standard environments. Here’s the archive version:
At #AAMAS2021, our researchers presented a reward-shaping #algorithm for #reinforcementlearning that works by trying to win against itself. We show that the modified algorithm converges faster & is more stable. Read here- https://t.co/4Lks9v5WNK
#research #researchpaper #scicomm
2
1
4
Charms of a catchy title! I’ve been guilty of the same … see this from a decade ago:
Beware, your date might need a lot of "attention" 👀
0
0
4
Key points: from the Stanford report - India has %>avg AI talent but <<avg publications. Reasoning remains a challenge for ML/DL in spite of GPT3, Copilot etc. - my bet is on neurosymbolic learning : but with distribution representations as ‘symbols’
Dr. Gautam Shroff, TCS Fellow, SVP & Head, TCS Research, @TCS recently shared valuable insights in his Keynote Session ‘Cutting Edge of AI from India for the World' at the #NASSCOMXperienceAI Virtual Summit 2021. Watch full video here https://t.co/9hyF4C8oZO
@gmshroff
@nasscom
0
2
7
Dr. Gautam Shroff, TCS Fellow, SVP & Head, TCS Research, @TCS recently shared valuable insights in his Keynote Session ‘Cutting Edge of AI from India for the World' at the #NASSCOMXperienceAI Virtual Summit 2021. Watch full video here https://t.co/9hyF4C8oZO
@gmshroff
@nasscom
0
6
11
Chap 11 of this excellent book by @kenneth0stanley should be required reading for all reviewers in AI conferences; and the whole book for anyone having to ‘manage’ research (like myself!).
scholar.google.co.in
KO Stanley, J Lehman, (No Title), 2015 - Cited by 266
1
6
36
SIGKDD India Chapter (@iKDD_News) is continuing the 'Data Science in India' event at KDD'21. Opportunity to know about Data Science ecosystem in India & connect with leaders from industry & academia. More at: https://t.co/5KxoxAftWl Register for free at: https://t.co/i7LCCDd1Em
0
10
15
Thanks @kdd_news for continuing hosting this event since 2015 with the KDD Conference. @iKDD_News @ravi_iitm @ManishGuptaMG1 @gmshroff @SharmaRaj
SIGKDD India Chapter (@iKDD_News) is continuing the 'Data Science in India' event at KDD'21. Opportunity to know about Data Science ecosystem in India & connect with leaders from industry & academia. More at: https://t.co/5KxoxAftWl Register for free at: https://t.co/i7LCCDd1Em
0
4
17
Large knowledge bases need to be used along with text corpora to train word embeddings as in this paper; nice work folks!
At #COLING2020, researchers from #TCSResearch presented a retrofitting-based model that incorporates relational knowledge present in knowledge graphs to improve static word #embeddings thus improving #NLP tasks such as #sentimentanalysis & #NER- https://t.co/ubMZXTDScl
#research
0
0
5
At #NeurIPS2020, researchers from #TCSResearch presented MetaNETs, an #AI-accelerated framework for designing artificial materials called #photonic #metamaterials which expedites the design and discovery of novel photonic devices - https://t.co/GZ8piSmLNs
1
6
25
This week, we continue to celebrate #WomeninResearch. Watch Senior Scientist Dr. Rekha Singhal, @Rekhasi27605481, talk about personal traits that have guided her to #research, and her fulfilling career at #TCSResearch. #inventingforimpact #womeninscience
3
8
14