ΑΙhub.org
deep dive
Defending against prompt injection with structured queries (StruQ) and preference optimization (SecAlign)
BAIR blog
06 May 2025
Recent advances in LLMs enable exciting LLM-integrated applications. However, as LLMs have improved, so have the attacks against them.
Copilot Arena: A platform for code
ML@CMU
28 Apr 2025
Copilot Arena is an app designed to evaluate LLMs in real-world settings by collecting preferences directly in a developer’s actual workflow.
Repurposing protein folding models for generation with latent diffusion
BAIR blog
14 Apr 2025
The awarding of the 2024 Nobel Prize to AlphaFold2 marks an important moment of recognition for the of AI role in biology. What comes next after protein folding?
Optimizing LLM test-time compute involves solving a meta-RL problem
ML@CMU
20 Jan 2025
By altering the LLM training objective, we can reuse existing data along with more test-time compute to train models to do better.
Generating a biomedical knowledge graph question answering dataset
Xi Yan
17 Jan 2025
Introducing PrimeKGQA - a scalable approach to dataset generation, harnessing the power of large language models.
Human-AI collaboration in physical tasks
ML@CMU
03 Jan 2025
Creating an intelligent assistant that uses the sensors in a smartwatch to support physical tasks such as cooking and DIY.
Improving calibration by relating focal loss, temperature scaling, and properness
Viacheslav Komisarenko
28 Nov 2024
In ML classification tasks, achieving high accuracy is only part of the goal; it's equally important for models to express how confident they are in their predictions.
AI in cancer research & care: perspectives of three KU Leuven institutes
This story is a collaboration of three Institutes that are working at the intersection of cell research, cancer research and care, and artificial intelligence.
Identification of hazardous areas for priority landmine clearance: AI for humanitarian mine action
ML@CMU
19 Nov 2024
In close collaboration with the UN and local NGOs, we co-develop an interpretable predictive tool to identify hazardous clusters of landmines.
Enhancing controlled query evaluation through epistemic policies
The winners of an IJCAI2024 best paper award explain the key advances of their work.
VQAScore: Evaluating and improving vision-language generative models
ML@CMU
06 Nov 2024
We introduce a new evaluation metric and benchmark dataset for automated evaluation of text-to-visual generative models.
No free lunch in LLM watermarking: Trade-offs in watermarking design choices
ML@CMU
23 Oct 2024
Common design choices in LLM watermarking schemes make the resulting systems surprisingly susceptible to watermark removal or spoofing attacks.
Linguistic bias in ChatGPT: Language models reinforce dialect discrimination
BAIR blog
30 Sep 2024
Examining how ChatGPT’s behavior changes in response to text in different varieties of English.
Rethinking LLM memorization
ML@CMU
23 Sep 2024
Our approach provides a simple and practical perspective on what memorization can mean, providing a useful tool for functional and legal analysis of LLMs.
How to evaluate jailbreak methods: a case study with the StrongREJECT benchmark
BAIR blog
09 Sep 2024
Providing a more accurate assessment of jailbreak effectiveness.
Proportional aggregation of preferences for sequential decision making
Nikhil Chandak and Shashwat Goel
27 Aug 2024
Read about work that won an outstanding paper award at AAAI 2024.
CMU-MATH team’s innovative approach secures 2nd place at the AIMO prize
ML@CMU
19 Aug 2024
Dive into our blog to discover the winning formula.
Are we ready for multi-image reasoning? Launching VHs: The Visual Haystacks benchmark!
BAIR blog
01 Aug 2024
This project focuses on the “Multi-Image Question Answering” (MIQA) task.
How to regularize your regression
ML@CMU
17 Jun 2024
Considering how to tune the norm-based regularization parameters in linear regression.
Beyond the mud: Datasets, benchmarks, and methods for computer vision in off-road racing
ML@CMU
17 Apr 2024
Off-road motorcycle racing poses unique challenges that push the boundaries of what existing computer vision systems can handle
Modeling extremely large images with xT
BAIR blog
08 Apr 2024
Introducing a new framework to model large images on contemporary GPUs while aggregating global context with local details.
The shift from models to compound AI systems
BAIR blog
15 Mar 2024
State-of-the-art AI results are increasingly obtained by compound systems with multiple components, not just monolithic models.
Unlocking the potential of entity-centric knowledge graphs: transforming healthcare and beyond
Christos Theodoropoulos and Leuven.AI Stories
27 Feb 2024
The concept of entity-centric knowledge graphs holds promise in reshaping how we organize, access, and leverage data.
On noisy evaluation in federated hyperparameter tuning
ML@CMU
12 Jan 2024
Our work explores key sources of noise and shows that even small amounts of noise can have a significant impact on tuning methods.
Asymmetric certified robustness via feature-convex neural networks
BAIR blog
14 Dec 2023
We propose the asymmetric certified robustness problem, which requires certified robustness for only one class and reflects real-world adversarial scenarios.
Goal representations for instruction following
BAIR blog
23 Nov 2023
How can we reconcile the ease of specifying tasks through natural language-based approaches with the performance improvements of goal-conditioned learning?
A comprehensive survey on rare event prediction
We review the rare event prediction literature and highlight open research questions and future directions in the field.
Test-time adaptation with slot-centric models
ML@CMU
25 Sep 2023
Improving out-of-distribution scene decomposition accuracy.
Training diffusion models with reinforcement learning
BAIR blog
15 Sep 2023
We show how diffusion models can be trained on downstream objectives directly using reinforcement learning.
A ‘black box’ AI system has been influencing criminal justice decisions for over two decades – it’s time to open it up
The Conversation
11 Aug 2023
Melissa Hamilton and Pamela Ugwudike investigate the use of automated decision-making systems in courts and prisons.
On the stepwise nature of self-supervised learning
BAIR blog
01 Aug 2023
Presenting a mathematical picture of the training process of large-scale SSL methods.
Navigating to objects in the real world
ML@CMU
24 Jul 2023
Research shows that modular learning is a reliable approach to navigate to objects.
Generating 3D molecular conformers via equivariant coarse-graining and aggregated attention
BAIR blog
14 Jul 2023
Introducing a variational encoder for molecular conformer generation.
GPT-4 + Stable-Diffusion = ?: Enhancing prompt understanding of text-to-image diffusion models with large language models
BAIR blog
26 Jun 2023
Our LLM-grounded model delivers improved prompt understanding in cases including negation, numeracy, and spatial relationships.
On privacy and personalization in federated learning: a retrospective on the US/UK PETs challenge
ML@CMU
05 Jun 2023
Studying the use of differential privacy in personalized, cross-silo federated learning.
TIDEE: An embodied agent that tidies up novel rooms using commonsense priors
ML@CMU
28 Apr 2023
We introduce a new benchmark to test agents in their ability to clean up messy scenes without any human instruction.
