Mansi Phute @ NeurIPS

Excited to be in Montreal this week for COLM 2025! I have two papers accepted this year: 1️⃣ The Unlearning Mirage, a dynamic framework for evaluating LLM unlearning. 2️⃣ With @Harsh_N_Lalai10, using 20 Questions as a creative framework to systematically evaluate geographical bias in LLMs. Check out the individual threads!

🚨 New paper alert! 🚨 Can you believe it? Flawed thinking helps reasoning models learn better! Injecting just a bit of flawed reasoning can collapse safety by 36% 😱 — but we teach large reasoning models to fight back 💪🛡️. Introducing RECAP 🔄: an RL post-training method that trains models to override unsafe reasoning, reroute to safe & helpful answers, and stay robust — all without extra training cost. ✨ Safer reasoning 🤖 ✨ Stronger jailbreak resistance 🔓 ✨ Lower overrefusal 🙅 ✨ Preserved core reasoning capability 🧠 #LLM #ReasoningModels #RLHF #AISafety #Alignment #MachineLearning

@ieeebigdata #AI #IEEE #BigData #DeepLearning #ResponsibleAI

🚨 NeurIPS 2024 🚨How robust are our AI-Generated Image Detectors? 🤔 Can they detect various magnitudes of image augmentations? 💡 Does performance fluctuate across scenes and subjects? 🚀 Find out with Semi-Truths: a dataset of about 1.5 million images for the targeted evaluation of AI-generated images. arxiv.org/abs/2411.07472

Presenting the all-woman team behind Semi-Truths (+ @PoloChau 😁), landed in #NeurIPS2024!! 📍Come see us tomorrow at 11am - 2pm, West Ballroom A-D #5211.

Introducing ClickDiffusion! We developed a system for precise image manipulation and generation that combines natural language instructions with visual feedback provided by the user through a direct manipulation interface.

🎉Thanks for the huge interest from the community in my last post comparing GPT-4o with UniTable for table recognition! We are excited to release more qualitative results of our UniTable!!! 😀 See more examples including paper and code in the thread 🧵(1/n)

Find the code here! 👉 github.com/poloclub/llm-s…

Looking to safeguard your LLMs from deception? 🛡️LLM Self Defense is 🛠️Simple: No fine-tuning or preprocessing 🌐Generalizable: Works for GPT3.5 and Llama 💪Effective: Countering almost all attacks Visit us at ICLR'24 tomorrow at Hall B #294 from 4:30pm - 6:30pm to know more!

Find the paper here! 👉 arxiv.org/abs/2308.07308

I’m honored to be awarded with the Marshall D. Williamson Fellowship for academic excellence and leadership by @gtcomputing 🌟 Thank you @PoloChau for your unwavering support and encouragement!

🤔 Want to Fine-Tune LLMs with Your Private Data for FREE? 3-in-1 tutorial designed for everyone! 📖 Accessible blog post medium.com/polo-club-of-d… + 💻 Step-by-step Colab notebook github.com/poloclub/Fine-… + 🎙️ High-level presentation

Thrilled to be donning the cap and gown for my Master's graduation today! 🚀🎓 I learned so many lessons at GT, both inside and outside the classrooms, that I will carry with me for the rest of my life. Thanks to everyone who supported me in this journey! Go Jackets! 🐝

Several CSE students, faculty, and alumni were recognized last week at @gtcomputing's 33rd Annual Awards Celebration! Stay tuned for our award wrap-up story to publish soon! Until then, here are a few highlights from the banquet! Congratulations, everyone! 🥳🎉

🎓 How to choose a PhD advisor? 💡My top tip: prioritize human qualities. Without the right people, research is not going to happen. 🌟Dive deeper here: medium.com/polo-club-of-d…

📢 Georgia Tech's School of Computational Science and Engineering has multiple faculty openings! cse.gatech.edu/faculty-hiring Join us at the coolest CSE department in the world! @GTCSE @gtcomputing @mlatgt

🛡️ Want to protect your AI models from attacks? You only need 3 robust principles! 🏆 #1 on RobustBench CIFAR-10 leaderboard 🔬 Robust Principles unlock the power of architectural robustness of models, datasets, adversarial training, and network designs! arxiv.org/abs/2308.16258

Congratulations to Ms. Mansi Phute (@mansiphute) for Wining #LIVELABZ Quiz on Practical Computer Vision. #LIVELABZ is a Knowledge Dissemination Initiative of Vishwakarma Institute of Technology Pune vit.edu/livelabz/ #AI #ComputerVision #Technology #MachineLearning

@ShripadBhat7 Fundamental and essential matrices are important in the epipolar geometry of stereo images. Both relate a point in one camera to a point in the other given rotation and translation. Essential matrix has 5 dof and fundamemtal matrix has 7 dof as it contains info about calibration

Q38. What is the Significance of Fundamental and Essential Matrix? Justify Their Relevance to Stereo Imaging #CVOQ

@ShripadBhat7 GMM is used to identify objects by classifying them in gaussian distributions. There are 2 stages in the model creation: expectation and maximization. In maximization we adjust the mean variance and weight of the distributions to maximize the possibility of correct classification

Q37. Why Weight, Mean and Variance are Updated in Gaussian Mixture Model? #CVOQ