Researchers from MIT and MIT-IBM Watson AI Lab have developed a tiny chip that can safeguard user data while enabling efficient computing on a smartphone. It provides a security solution for power-hungry AI models, offering protection against two common attacks.
news.mit.edu/2024/tiny-chip…

When a human-AI conversation involves many rounds of dialogue, the powerful large language machine-learning models like ChatGPT sometimes start to collapse. A team of researchers from MIT and elsewhere has pinpointed a surprising cause of this problem. mitsha.re/KYxV50Rk5h5

Congratulations to MIT and MIT-IBM Watson AI Lab researcher William Oliver and his colleagues for their election as fellows to the AAAS!
news.mit.edu/2024/four-mit-…

Last-mile logistics is a tricky combinatorial optimization problem for operations research specialists. The MIT-IBM Watson AI Lab's Matthias Winkenbach and his team are developing AI methods to make vehicle routing more efficient and adaptable for unexpected events.
news.mit.edu/2024/matthias-…

We have a cool challenge on understanding document images in our 2nd #CVPR2024 workshop on “What is Next in Multimodal Foundation Models?”, (sites.google.com/view/2nd-mmfm-…). This is a great opportunity to showcase your work in front of a large audience (pic below from our 1st workshop)

A new technique teaches a large language model to perform a safety check on an AI chatbot so it doesn’t give toxic responses. “Our work is an attempt to reduce the human effort to ensure a safer and trustworthy AI future,” Pulkit Agrawal says. mitsha.re/46Jp50RcfRg

Ever asked yourself what's the best explainability method for ViT Transformer at the moment?

🚨 We present you LeGrad, a Layerwise Explainability GRADient method for large ViT transformer architectures. 🚨

So, grab your ☕️ and 🥐, release your inner 🇫🇷, and join us on our…

Tamara Broderick uses statistical approaches to understand and quantify uncertainty in data, which can affect study results. “The reality is that we live in a noisy world, and we can’t always get exactly the data that we want,” she says. mitsha.re/HXoG50QJVc4

Hey, we did a thing: 'LAB: Large-scale Alignment for chatBots'—a new synthetic data-driven LLM alignment method that yields great results without using large-scale human or proprietary model data.

arxiv.org/abs/2403.01081
models: huggingface.co/ibm/labradorit…, huggingface.co/ibm/merlinite-…

Congratulations to Dean Chandrakasan on being name MIT's inaugural chief innovation and strategy officer, in addition to his continued role as dean of engineering. news.mit.edu/2024/anantha-c…

MIT-IBM Watson AI Lab PhD interns look to improve natural language usage by investigating AI reasoning and reliability.
news.mit.edu/2024/reasoning…

Generative AI meets physics-based simulation to create robots for physical tasks: bit.ly/48HZjml

Our team is currently looking to hire PhD candidates for 2024 summer internship, to work on efficient training and inference of large language (and/or multimodal) models. Please DM or send me an email if you are interested. MIT-IBM Watson AI Lab IBM Research

Working with the MIT-IBM Watson AI Lab, master's students Irene Terpstra ’23 and Rujul Gandhi ’22 use language to design new integrated circuits and make it understandable to robots. news.mit.edu/2023/leveragin…

A flexible solution to help artists improve animation: This new method draws on 200-year-old geometric foundations to give artists control over the appearance of animated characters. mitsha.re/AHkH50QkGxK

A customized onboarding process from the MIT-IBM Watson AI Lab helps a human learn when a model’s advice is trustworthy, improving human-AI collaboration.
news.mit.edu/2023/automated…

Over 2,000 years ago, the Greek mathematician Euclid, known as the father of geometry, changed how we think about shapes. Now Justin Solomon is applying modern geometric techniques to problems in computer vision, machine learning, and statistics. mitsha.re/Hw5h50QjaSw

⚡️Energy Transformer (ET)⚡️ A novel architecture combining 3 prominent ideas in AI

1️⃣ Transformers: mix tokens with attention
2️⃣ Energy-based Models: inference descends a tractable energy function
3️⃣ Associative Memory: inference performs error correction

#NeurIPS2023

A 🧵:…

Benedetto Marelli develops silk-based technologies for helping crops grow, preserving perishable foods, and more. “Knowing that the technology can move the needle in crises such as food waste is the highest reward of all,” he says. mitsha.re/QJLm50Qf3oP

Now that the CVPR deadline has passed, it's a good time to share our new #EMNLP23 work. In this paper, we show one possible solution to the 'bag of objects' behavior of VL models' representations, which tend to ignore object attributes and relations.

arxiv.org/abs/2305.06343