Mrigank Raman

NLP enthusiast, MSML@CMU | Applying to PhD programs this cycle

[Mrigank Raman @MrigankRaman]

🚨⚠️ Stop using the [CLS] token ⚠️🚨

I will be talking about 1 simple trick to astonishingly boost the robustness of your NLP classifers.

Today, 2pm at #EMNLP2023

"Model-tuning Via Prompts Makes NLP Models Adversarially Robust"

📝https://t.co/lEFG1DVoWh

Summary below 1/🧵

[Pranav @PranavMani30]

Does adapting general-domain models to medical-domain actually help w med-domain tasks?
Stop by at Tuttle Hall, 230p EST, Nov 14 @emnlpmeeting to catch the amazing @danielpjeong present his 🚀oral 🚀talk.
Super glad to be part of this work w @danielpjeong @saurabh_garg67 @zacharylipton @MichaelOberst
Paper: https://t.co/wy4WQjxsrT

[Daniel P Jeong @danielpjeong]

🧵 Are "medical" LLMs/VLMs *adapted* from general-domain models, always better at answering medical questions than the original models?

In our oral presentation at #EMNLP2024 today (2:30pm in Tuttle), we'll show that surprisingly, the answer is "no".

https://t.co/3259JyNU44

[Pratyush Maini @pratyushmaini]

Our work on ACR memorization won the Best Paper Award at CONDA @ #ACL2024 🎉🎉
I will be giving a talk on the same on August 16th

Over the last few months, I have substantially grown in the realization of how impactful ACR will be in the GenAI copyright discourse. Thoughts🧵1/n

[Pratyush Maini @pratyushmaini]

I am at ICLR 🇦🇹 all week presenting some recent works with my collaborators. Find me at one of these places 👇 or drop a DM/e-mail if you would like to chat! I am excited to talk about data quality, memorization, and life @datologyai! Looking forward to meeting new people!

[Pratyush Maini @pratyushmaini]

I am super excited to chart this new journey with an incredibly talented team!! We are on a mission to bring research from labs🔬 to the real world 🌎 and define new frontiers for high quality data research!

[Abridge @AbridgeHQ]

📣 Today, we are announcing our $150M Series C! This is one of the largest funding rounds made to date in generative AI for healthcare.

We are grateful to the luminary investors and iconic institutions who believe in us and our mission—to power deeper understanding in healthcare. This round was led by @lightspeedvp who will be joining our board. Other new and existing investors joining the round include co-lead @Redpoint, with support from @IVP, @sparkcapital, @usv, @BessemerVP, @WittingtonVC, Mass General Brigham Artificial Intelligence and Digital Innovation Fund (AIDIF), Kaiser Permanente Ventures, and CVS Health Ventures.

With this new capital, we will continue to push boldly into fundamental research, developing bedrock foundation models that draw upon vast troves of multimodal healthcare data.

Looking forward to this new wave of possibilities as we continue to fulfill our vision of improving the lives of patients and clinicians.

Read more about our Series C here: https://t.co/cYC4Dt6iae

[Pratyush Maini @pratyushmaini]

1/4 The Right to be Forgotten is knocking on the door. Yet, unlearning in LLMs has no clear task definition, no evaluation metrics or baselines. Introducing TOFU: Task of Fictitious Unlearning for LLMs
🌐 https://t.co/XYXOOxd3lo
w/@A_v_i__S @zhilifeng @zacharylipton @zicokolter🧵

[Mrigank Raman @MrigankRaman]

🎊🎊Submitted my 10th and final PhD application for this cycle🎊🎊

Submitting applications while traveling to EMNLP and NeurIPS was a very hectic albeit cathartic experience.
Hoping for the best🤞

A big thank you to @zacharylipton, @danish037 @LiangDavis for providing me LoRs

[Saurabh Garg @saurabh_garg67]

Does contrastive pretraining on diverse data, give models robust to distribution shift?

Spoiler: Better than ERM but there is a _huge_ room to improve, e.g., with pseudolabeling

📝: https://t.co/4b3GOysXks

w @setlur_amrith @zacharylipton Siva B. @gingsmith @AdtRaghunathan

1/

[Pratyush Maini @pratyushmaini]

Been an amazing week meandering through the poster lanes at #EMNLP2023. ICYMI, here are some of my favourite posters on memorization/contamination, data quality, and trustworthiness.

[Hariharan @manik_hariharan]

Without finetuning, can LLMs learn from tabular data?

We show that LLMs can perform tabular reasoning through automatically generated prompts, and such prompt-based learners work inside a boosting procedure.

#NeurIPS2023 12/12 6.15pm EST #1915. @yidingjiang @zicokolter

1/

[Mrigank Raman @MrigankRaman]

@WilliamWangNLP Oral papers should also have a poster session imo like ML conferences

[Tanya Marwah @__tm__157]

The architecture design space for neural net-based #PDEs solvers is still very under-explored. In our #NeurIPS2023 paper we study the benefits of weight-tied architectures for steady-state PDEs.
Paper: https://t.co/EDIaIqRZi6
Code: https://t.co/vMrFfdNBZ9

[Sachin Goyal @goyalsachin007]

Mrigank is an outstanding budding researcher and is looking for grad school opportunities (PhD) this season!

[Pratyush Maini @pratyushmaini]

Encoder-only classifiers are the backbone of critical NLP applications: toxicity classifiers, GPT detectors & data-quality filters. These applications demand OOD & Adv robustness.
We have all been fine-tuning them wrong❗
Come talk to us at #EMNLP2023 & robustify your classifiers

[Mrigank Raman @MrigankRaman]

This work was the fruit of a wonderful collaboration w/@PratyushMaini @ZicoKolter @ZacharyLipton @danish037

👨🏻‍💻 https://t.co/oXQJuVdV3r

7/7

[Mrigank Raman @MrigankRaman]

🚨⚠️ Stop using the [CLS] token ⚠️🚨

I will be talking about 1 simple trick to astonishingly boost the robustness of your NLP classifers.

Today, 2pm at #EMNLP2023

"Model-tuning Via Prompts Makes NLP Models Adversarially Robust"

📝https://t.co/lEFG1DVoWh

Summary below 1/🧵

[Mrigank Raman @MrigankRaman]

Our work suggests practitioners use MVP to fine-tune pre-trained NLP models, irrespective of the data size (few-shot or full data), architecture (encoder/decoder) and model capacity (small or large).

6/🧵