
PhD student at Berens lab, HertieAI @uni_tue ELLIS PhD student & ass. IMPRS-IS | Ex @fz_juelich @KTHuniversity @IISERPune ML for science. NeuroAI.
Joined March 2020
- Tweets301
- Following1.5K
- Followers352
- Likes4.7K
Harini Sudha retweeted
feeling a bit under the weather this week … thus an increased level of activity on social media and blog: kyunghyuncho.me/i-sensed-anx…
Harini Sudha retweeted
#Neuromorphic #computing just got more accessible! Our work on a Neuromorphic Intermediate Representation (NIR) is out in @SpringerNature Communications. We demonstrate interoperability with 11 platforms. And more to come!
nature.com/articles/s41467-0…
A thread 🧵 1/5
Harini Sudha retweeted
Let's be honest. For everyone in ML/NLP, it's a really exciting time but also very stressful with so many new papers, models, benchmarks, deadlines, reviews, talks, workshops, conferences...
How do you keep up and stay sane?
For me, the only solution is collaboration. 1/n
Harini Sudha retweeted
There appears to be a mismatch between publishing criteria in AI conferences and "what actually works". It is easy to publish new mathematical constructs (e.g. new models, new layers, new modules, new losses), but as Apple's MM1 paper concludes:
1. Encoder Lesson: Image resolution has the highest impact, followed by model size and training data composition.
2. Vision-Language (VL) Connector Lesson: Number of visual tokens and image resolution matters most, while the type of VL connector has little effect.
3. Data Lesson 1: Interleaved data is instrumental for few-shot and text only performance, while captioning data lifts zero-shot performance.
4. Data Lesson 2: Text-only data helps with few-shot and text-only performance.
5. Data Lesson 3: Careful mixture of image and text data can yield optimal multimodal performance and retain strong text performance.
6. Data Lesson 4: Synthetic (caption) data helps with few-shot learning.
I suspect it would be very hard to publish a paper that says "we made the images bigger and got better results".
There is great value in careful ablations, scaling studies (as this paper shows when determining learning rates), data science, data engineering and engineering in general. Huge work goes into constructing the data pipelines.
Engineering for a long time had a poor reputation in AI conferences - many papers were rejected with "this is just engineering! ". In the end, just engineering is working beautifully. There is room for scientific invention, but just engineering also allows for an incredible amount of innovation.
arxiv.org/pdf/2403.09611.pdf
Harini Sudha retweeted
Being in the AI research/industry space today is exhausting - every day, you open Twitter and, even if you work on this every day, you still get massive FOMO.
Harini Sudha retweeted
🚨 2 Days Left to Join our Networking Event! 🚨 buff.ly/4aOqDjY
🔥 Plus, get ready for our Workshops! Present your work in <1 min & dive into discussions on climate effects on the brain & #greenwashing: buff.ly/3TXvNEv
#Neuroscience #ClimateScience #Networking
Super happy to be accepted to @scioi_cluster summer school on #Embodied #Intelligence at Berlin, late August! I'm excited and looking forward to discussing my master thesis research during the poster session. Also, curious about various perspectives of building intelligent system
13
1,087
Harini Sudha retweeted
Proof that a basic Biology neurobio training @IISERPune prepares one for much more. #neurobiology #biology to #neuromorphic #computing testament to #interdisciplinary training and the joy of learning of the student population. #teaching and #research made to work
A tweet after a long time. But, it's worth mentioning here.
I defended my Master thesis yesterday doing a project on neuromorphic computing where I leverage the benefits of event cameras and neuromorphic hardware to identify some low level features. (1/2)
Learnt a TON at the NeuroComputing Systems lab @KTHuniversity and always gonna be super grateful for the mentorship! Coming from a pure science background, it was fun working at junctions of EECS and neuroscience. Looking forward to doing more cool stuff in the field! (2/2)
8
729
Harini Sudha retweeted
Replying to @ylecun
“The first principle is that you must not fool yourself—and you are the easiest person to fool. So, you have to be very careful about that. After you’ve not fooled yourself, it’s easy not to fool other scientists. You just have to be honest in a conventional way after that.” — Feynman
Harini Sudha retweeted
Replying to @curiouswavefn
J. Robert Oppenheimer's letter of recommendation for Richard Feynman for a position in the Physics Department at the University of California, Berkeley.
"He is the second Dirac, only this time human.”
Harini Sudha retweeted
ChatGPT for Research
No more sifting through endless research papers to find the answer you need.
Just ask any question and Consensus AI will deliver a GPT-4-powered summary of the top 5-10 papers.
No wonder people already have research papers worth a PhD before applying for a PhD.
In mid 2000, when I became an assistant professor, I used to get research request from prospective PhD students and masters projects, Then started bachelors thesis and summer projects. Now i am getting such requests from class 10 students. Wondering what to expect next!!
1
312
Harini Sudha retweeted
In college, a journalism professor gave me the best advice when it came to interviewing/getting to know people:
"No one's inherently boring. If they're boring, it just means *you* haven't asked them the right questions yet."
Harini Sudha retweeted
Life takes you places you never expect.....!
Article penned
,thanks to @IndiaBioscience
for giving me this opportunity ,special thanks to
@OSNVancouver @silents0up
@YoursHarini @Friendzymes @iGEM
indiabioscience.org/columns/…
Harini Sudha retweeted
🎓 Ready for the 7th HBP Student Conference #HBPSC2023? The programme has been updated and you can now find information about the possible workshops and our keynote speakers online!
Take a look below and while you are at it - don't forget to register! 👇
humanbrainproject.eu/en/educ…























