Rohan Pandey
@rohan99pandey
Followers
363
Following
294
Media
1
Statuses
314
Robustness, NLP & Healthcare | CS PhD @UMassAmherst | Prev. @mckinsey @ShivNadarUniv
Joined October 2023
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Hamas
• 924064 Tweets
#NXTNoMercy
• 84114 Tweets
ESPN
• 73635 Tweets
YouTube TV
• 58041 Tweets
Dinamarca
• 49939 Tweets
Big 10
• 35792 Tweets
Giulia
• 34639 Tweets
Falcao
• 24855 Tweets
Q33 U STEAL MY HEART
• 22293 Tweets
DirecTV
• 20982 Tweets
Roxanne
• 17774 Tweets
Hudson
• 17538 Tweets
Braves
• 16569 Tweets
MIX師
• 15850 Tweets
Grant Justice
• 15084 Tweets
Trojans
• 10987 Tweets
Lacy
• 10953 Tweets
解雇規制緩和
• 10692 Tweets
Jaida
• 10498 Tweets
Pinned Tweet
1/7 As a CS PhD student, I've often reflected on how challenging it can be for newcomers to enter research, especially with limited mentorship. The questions, uncertainties, and lessons I've encountered inspired me to create a resource I wish I had when starting out.
1
31
406
7/7 If you're interested in CS research or know someone who might be, check out the full guide:
I welcome any feedback or questions – this is a learning process for me too!
4
12
120
I am excited to announce that I successfully survived my first year in a PhD program. One of the biggest takeaways for me, is that there is more to my identity than being a PhD student (not as easy as it sounds!).
6
1
74
2/7 So, I tried to write: "A Beginner's Guide to Getting Started in CS Research" - a series of documents aimed at demystifying the research process for those curious about or just entering the field of Computer Science research.
1
2
41
@aidan_mclau
I would argue that meta-meta-learning is even more important... and so on. learning to learn to learn to learn...
4
1
35
3/7 What is it? The first part of this series provides a comprehensive overview of the CS research process, broken down into four main phases:
- Consumption
- Ideation
- Experimentation
- Presentation
1
0
27
@kdexd
As someone just starting out in a PhD, it is absolutely difficult to bog down and zone on in what is meaningful and what matters. Extremely overwhelming and alarming to see the rush which is really making me not wanting to work on LLMs at this point.
1
0
19
1/9 Fed up with all the AGI hype, but don’t know where to start? That's why I’ve been putting together a beginner-friendly guide series to help democratise access to CS research!
2
0
19
@radshaan
not necessarily, it only works if you have a certain amount of money... try going into any of the public healthcare systems. I have been brought up in the biggest public hospital, and it's a wreck for 99% of the population. (it is primarily a population problem)
1
0
17
5/7 Who is it for? This guide is meant for anyone curious about CS research:
- Undergrads considering grad school
- Professionals thinking about academia
- Anyone with limited access to research mentorship
It assumes little to no background in research.
1
0
19
4/7 Drawing from my (little to none) experiences as a research student focusing on robust language models in healthcare, I've tried to offer practical insights into each of these phases and address common questions newcomers often have.
1
0
18
6/7 The goal? To provide a starting point for those interested in CS research, offering a high-level view of the process and answering questions like "When do I know enough to start?" and "How do I organize what I've learned?"
2
0
16
@alfcnz
Try . Extremely well detailed, funny and also has a paid app that plans everything for you if that’s what you need
1
0
12
@akshay_pachaar
It would be great if you could attribute the original creators, and people in such posts in the first tweet! It goes a long way for people putting in hours to build such useful resources.
0
0
11
@MicrobiomDigest
In my previous life in consulting and now in research, 1) I make sure everyone gets a chance to speak (actively encourage everyone), 2) Go out of my way to mentor underrepresented people, 3) Just be available for chats/ideas/work.. It all begins at an individual level.
0
1
10
My take: If you're a student: work on problems you care about (and don't be married to a method) with people you enjoy working with.
If you're a student wanting an exciting life, good comp, and impact on the real world: work on LLMs
32
20
569
0
0
7
@Ana_BurgosRuiz
Excellent idea! Although a lot of the collaborations I see come from either a very grant based collaboration or prestige (onboarding a senior author). If you start looking you'll see the trends -- but hopefully this helps. A lot of open communities like
@CohereForAI
work!
0
0
6
I've just started my PhD in machine learning and I almost feel compelled to work on LLMs (talk about fomo). As someone exploring/figuring out areas of interest, would love to hear about other unexplored open problems worth investigating!!
#AcademicTwitter
#NLP
#MachineLearning
4
0
5
@NC_Renic
I feel incredibly lucky to have all of my close undergrad friends - doing completely unrelated things. Our conversations usually are just banter and enjoying each other's existence, the most refreshing bunch of folks!
0
0
5
ML twitter: What opinions do we have on quantity of paper's individuals (grad students/profs) are publishing every year. When I see grad students publishing 10+ papers, I'm immediately skeptic.
2
0
5
@rajammanabrolu
This makes me wonder more about what is the time to do these optimisations. For eg: is undergraduate the right time? If yes, then even that is biased towards people from certain communities that already know the game in some sense.
1
0
5
@jxmnop
It should be super good for practice! Hope this doesn't become the ONLY hiring signal for research tho. Maybe for engineering might make sense.
1
0
5
@_jasonwei
I believe you are describing a move from a role where you are a part of many, to a place where you have to own components E2E (also difference in roles from an IC to a PM/PO). Both have merits and one could be fulfilling for an individual than the other (which is what matters).
0
0
0
@aryaman2020
@saurabh_shah2
Nope, not necessarily, for me, it's a way to buy some time in life, that allows me to explore (my curiosity) and do things in the interestingness directions. I don't care what comes out of this, I'm here for the ride.
1
0
3
@aryaman2020
@saurabh_shah2
I decided to let go of a very cushy job in India, in a decent tech consulting setup, just because I wanted to reduce the "regret" in the long term.
Is my PhD my entire identity? Absolutely not. I'd rather be known for my personality, than my research in the long term.
0
0
4
@divy93t
I agree with the RA positions as a great way to getting started with research, but there are two caveats with it, a) underpaid compared to other industry positions, and b) preference to students with certain backgrounds (often just a numbers problem to filter applicants).
1
0
4
It's in the name 🔥🔥
📢 Excited to finally be releasing my NeurIPS 2024 submission!
Is Chinchilla universal? No! We find that:
1. language model scaling laws depend on data complexity
2. gzip effectively predicts scaling properties from training data
As compressibility 📉, data preference 📈.
🧵⬇️
17
114
825
2
0
4
9/9 Check out the complete second edition here: and also on my website . PS. This document has been proudly been written without any LLMs (unlike this thread).
0
0
4
The UMass CICS PhD Application Assistance Program (PASP) is now open! Current students will provide prospective candidates from under-represented communities feedback on their application materials.
Follow Link:
2
2
4
@Skiminok
@karpathy
@DynamicWebPaige
@Stefania_druga
@moyix
@oegerikus
Democratising access to knowledge (in all fields) is truly one of the greatest gifts of this AI revolution.
0
0
4
@nealkhosla
disagree, 1) it has high value in relocating to locations that you wouldn't otherwise get a chance to, 2) good experience for people that do not get quality undergraduate education, 3) (underrated) easier to meet people / make friends when in school vs moving for a job.
0
0
3
@giffmana
100% agree, the word that comes to my mind is “immersion” - married to the process and art.
0
0
2
@jxmnop
It is insane how a lot of it feels so much more like a marketing game at this point! Academia has started feeling more like a startup competition at this point!
0
0
3
@saurabh_shah2
@khoomeik
@ElytraMithra
@sahitYAML
We've been talking about it for ages now
@khoomeik
-- sending you a text.. time to get cracked fr. We've got to name the paper PANDEY some how
0
0
3
@BasedBeffJezos
@teortaxesTex
Or there is no meta-point here, I'd just argue he's just a guy who wants to improve life expectancy and loves science. I personally don't like the cultish worldviews (no offence) towards everything lol
3
0
3
@thegautamkamath
It's a part of the zero-sum game right, a lot of the grants, best-papers etc (accolades) are limited. I personally do not care too much for these (I am also a grad student) and really celebrate the success of my peers! We're all fighting the same fight.
0
0
3
64 A100s for a research group is 🤯🤯, although to be fair stanford NLP does solid work and deserves more. If anyone is looking to donate, happy to be a taker too! (even one would be great!)
@animesh_garg
@drfeifei
@chrmanning
@percyliang
Sad, but not quite true…. Really it’s that we’ve got 64 A100s. We’ve got 100+ other GPUs too.
Still … not very many. The main thrust of the argument is valid.
If anyone would like to give us some more GPUs, do get in touch!
5
9
127
1
0
2
In order to bring drip back to academia, I am committing to wearing a suit for my first PhD paper presentation, if/when that happens! A fitting ode to my prior consulting life. Open to other ideas too 🧐
1
0
3
2/9 This is Part 2 of my "How to Get Started in CS Research" series. Why am I doing this? Because I remember that time I tried to read a research paper, and thought I was reading an alien language.
1
0
3
@miniapeur
Yeah I feel this more and more now, as I grow older - the reality of change in so many variables is bittersweet!
0
0
3
@giffmana
I agree, but I have seen increasing rewards with building in public, and especially sharing "not so SOTA" results or anything interesting for that matter. As for academic lineage, I try not to think about things beyond my control!
1
0
3
@jeremyphoward
I personally like to think of the question "how to decide what is worth doing" as a grad students as a function of i)interest, ii)Uniqueness (of your lab/position) iii) difficulty -- this usually is a good heuristic for me.
0
0
2
@Teknium1
I don't get the hype around this "intelligence" conversation - just make them work for tasks we care about and keep making them better. .
0
0
3
@BenjiXie
I just did a quick glance, but it seems extremely useful and detailed! Bookmarking this - in a similar spirit I tried doing a similar breakdown not for applying to PhD's but doing research in general.
1/7 As a CS PhD student, I've often reflected on how challenging it can be for newcomers to enter research, especially with limited mentorship. The questions, uncertainties, and lessons I've encountered inspired me to create a resource I wish I had when starting out.
1
31
406
0
0
1
@yoavgo
All my papers in ARR have been a similar experience, it starts with the title and abstract which sets up high expectations, and by the end it's very disappointing.
0
0
2
@indygupta
@tianyin_xu
Interesting, do you think we as a community can also do better with our information dissemination? I know there is little incentive to us writing "layman blogposts" or "how is it useful for you?" guides -- but I've always thought we're missing that.
2
0
2
@khoomeik
Broadstrokes I think if it's templatic it isn't really very interesting.. The compute friendly approach I feel is start with toy/synthetic datasets - on the problem you care about - find "real world" datasets and verify (also try to prove bounds if you can).
1
0
1
@DavidSKrueger
@RolandMemisevic
I'm not surprised, a lot of what I see nowadays is just re-packaging old stuff with a newer umbrella term, which allows people to use the current "hype" in their favour (read: AGI, Alignment,....)
I do agree in your case it's a bit more blatant lol...
0
0
1
@saurabh_shah2
Yeah, I think people mistakenly draw the equivalence of life == work (for everyone). I have seen so many people who treat their work as just work, and lead the most loving and exciting personal lives.
2
0
2
@MountainOfMoon
This just makes me wonder, what is the right time for getting kids zoned into a particular area. For comparison, even in sports, a lot of them are put into certain sports at a very young age and go all in. I guess it also is very subjective
1
0
2
@giffmana
Appreciate you calling things out publicly, useful for people looking at things for the outside to get all perspectives
0
0
2
3/9 Continuing from the previous document, this guide breaks down the process into manageable steps, focusing on the often-overlooked "consumption phase" or literature review (to avoid working on something that has already existed for the last 10 years!)
1
0
2
@JubaZiani
Congratulations! I often think people forget the importance good teaching can have on an individual, and really change their future!
0
0
2
I have a few hypothesis, 1) large compute -> grid search through possibilities, write what works, 2) hierarchy that works, postdocs>phds>masters>undergrads, 3) lot of low hanging fruits, 4) lots of collaborations/low bar for inclusion in papers, 5) salami slicing gone worse?
2
0
2
1
0
2
@abeirami
@ReviewAcl
I agree, I received a "If all reviewers increased their load by 3 then we would cover our needs." -- increasing my load by 3 is more of than not impossible without decline in review quality.
0
0
2
@WilliamWangNLP
We really need a stronger and more organized community of AI researchers, likely in a format, that argues for open science and pulls the conversation away from the AI doomers towards a more centred approach.
1
0
2
6/9 Struggling with information overload? I discuss how I effectively search for papers, organize my findings, and develop a broad understanding of a topic without getting lost in the details.
2
0
2
@radshaan
partly, yes you do have a "5-star" hospital experience that you can get, but also you get pointless surgeries/medications/tests that you never need. A lot of them have "billing targets" internally. I'd imagine it be better in certain countries (not sure tho)
1
0
2
@kchonyc
modern LM research is a lot more about marketing, and a lot of early researchers have to play the game or they fall off. Do you think there are better incentive mechanisms that align the personal objectives (citations,papers,roles) with the broader objectives (good for all)?
0
0
2
0
0
2
@jmhessel
It's always about the people. It's easier to work on things you don't like with amazing people vs working on things you like with terrible people.
0
0
2
@m2saxon
It's also surprising why no one is talking about this? I mean I see it everyday, especially bigger groups, churning papers where the senior authors couldn't possibly have "worked" on the 50+ papers they published in a year. It's like a CEO putting their name on every presentation
1
0
2
8/9 Whether you're a student starting your first research project or a professional looking to enter academia, this guide provides valuable insights to help you navigate the CS research landscape.
1
0
2
Good reminder of getting the basics right, there's a lot of benefit in starting small and working up to complicated searches (if required).
It can't be repeated enough: learning-rate is the single most bang-for-buck thing you can tune.
If you think you know *ze best* learning-rate, it just means you only train standard stuff!
This is not a "secret trick" either; it's stated very clearly in THE deep-learning book:
19
46
546
0
0
2
@zacharylipton
@HochreiterSepp
@davekale
Thank you for sharing this, as someone who's just starting his PhD - this serves a healthy reminder of enjoying the small moments (and life!)
0
0
2
@wzhao_nlp
4. I have a 2x more specific and constructive feedback than other reviewers (where it looks like they barely put in an hour on each paper)
1
0
2
@AntoniaKaramol
I'm trying to hypothesize the various reasons: 1) it could senior authors who are running big labs. Which i guess is ok?(not sure) if written over decent time frames and contribution criteria met. or 2) the rigour of each paper has gone low, easier to slice up papers?
0
0
1
@giffmana
Also, I would argue and say don't write a paper for things that could be a blogpost. Write it in the same spirit but making it more accessible might be more useful in the long term.
1
0
2
Often when I'm reviewing a clearly unpolished work, I still take the time to write detailed feedback and improvement ideas. While every other review is mostly a few sentences long. Am I missing something?
0
0
2
@indygupta
@tianyin_xu
I agree but realistically, there is a lack of incentives for non-research teams have to read papers. 1) The barrier to reading through artifical complexity in research papers, 2) If I say I want to spend 2 hours of my work time reading papers, I'd be told to write more code lol
2
0
2
@giffmana
100%, that's also a general choice I think, do things 100% vs half-assing. And the rigor, I could argue, people could also skimp on with pre-prints lol
0
0
2
@MLStreetTalk
@cohere
@nickfrosst
100% agree, the only correction I would make is let's make models solve (not just business) real world problems
0
0
2
@_vaishnavh
@GregorBachmann1
One of the most interesting works I have seen in a while! Thanks for sharing.
0
0
2
@miniapeur
I started waking up early, working out heavy, physically tiring myself out by the end of the day.. has allowed me to better my sleep. Although I still wake up in the middle of the night problem solving in my head, but it's gotten better. Zinc + Magnesium help
0
0
2
@jbhuang0604
David Jensen teaches an excellent class for empirical research methods at UMass which overlaps with your content, I would recommend also reaching out to him! It’s one of the best classes I have taken partly due to him being an excellent lecturer
0
0
1
@npparikh
I find academia interesting because of the nature of mentoring and working with brilliant minds. I have worked in consulting before, which I also liked. At the end, I feel it is about picking your poison and enjoying what you are doing, rather than constantly wondering
0
0
2
0
0
2
7/9 There’s also a lot more FAQs (than last time) addressing questions like:
"Should I follow the hype or my heart in research?"
"How do I know when I've read enough?"
"What makes a research area 'interesting' (besides the cool factor)?"
1
0
2
5/9 One key takeaway: Know yourself! The guide helps you identify whether you're more inspired by real-world problems (top-down approach) or specific technical challenges (bottom-up approach).
1
0
2
@abeirami
@EugeneVinitsky
But sometimes even those are useful, in a particular setting X works for me vs Y in another setting!
0
0
2
@maksym_andr
@ncmeade
Yeah I think trying to dis-entangle which part of the procedure is causal is extremely tough. The pythia framework by
@BlancheMinerva
is fantastic for running such studies!
0
0
2
@khoomeik
tbh i just don't like the cult-y vibes of the community... sell me an important problem, not a "this solves it all" technology (cause there is none).
1
0
2
4/9 This edition detailed steps covers four main steps:
0️ Finding your area of interest
1️ Discovering relevant papers
2️ Understanding the research landscape
3️ Diving deeper into specific topics
1
0
2
0
0
1
@lunwang1996
Anything on twitter, moves into my zotero "interesting reads", and I just prune those over time. Any topic specific reads have their own space.
0
0
1
@RamanDutt4
@turinginst
@BioMedAI_CDT
@tisimpson
@doyarzunrod
@tmh31
@STsaftaris
Congratulations brother! keep pushing as usual!
1
0
1