International Conference on Learning Representations

A new method called self-disciplined autoregressive sampling (SASA) allows large language models to detoxify their own outputs, without sacrificing fluency.

MIT researchers find large language models process diverse types of data, like different languages, audio inputs, images, etc., similarly to how humans reason about complex problems. Like humans, LLMs integrate data inputs across modalities in a central hub that processes data in an input-type-agnostic fashion.

MIT and other researchers developed a framework that models irrational or suboptimal behavior of a human or AI agent, based on their computational constraints. Their technique can help predict an agent's future actions.

A new computational approach developed at MIT makes it easier to predict mutations that will lead to optimized proteins, based on a relatively small amount of data.

A new multimodal machine-learning technique from the MIT-IBM Watson AI Lab blends two kinds of self-supervised learning methods to learn more similarly to humans.

Self-play has helped AI systems succeed in games like chess and Go. Can the same method help improve AI programming abilities? Using easy-to-check, hard-to-solve programming problems, researchers show AI can create, solve, and train on its own puzzles:

Chemical engineers and materials scientists are constantly looking for the next revolutionary material, chemical, and drug. The rise of machine-learning approaches is expediting the discovery process, which could otherwise take years. "Ideally, the goal is to train a machine-learning model on a few existing chemical samples and then allow it to produce as many manufacturable molecules of the same class as possible,…

MIT researchers have developed a technique to train a machine-learning model for image classification, which does not require the use of a dataset. Instead, they use a "generative model" to produce synthetic data that is used to train an image classifier, which can then perform as well as or better than an image classifier trained using real data.

A new large language model framework teaches LLMs to use an optimization solving algorithm to resolve complex, multistep planning tasks. With the LLMFP framework, someone can input a natural language description of their problem and receive a plan to reach their desired goal.

A team of researchers at Edinburgh University tested some top multimodal large language models to see how well they could answer questions based on images of clocks...

Language models are taking a significant step forward in understanding and executing abstract concepts.

A new machine-learning technique can train and control a reconfigurable soft robot that can dynamically change its shape to complete a task. The researchers, from MIT and elsewhere, also built a simulator that can evaluate control algorithms for shape-shifting soft robots.

A new technique can more effectively perform a safety check on an AI chatbot. MIT researchers enabled their model to prompt a chatbot to generate toxic responses, which are used to prevent the chatbot from giving hateful or harmful answers when deployed.

The technology still needs more work before it is ready to be deployed at scale.

Apple researchers have released a new open-source AI model that is capable of editing images based on a user's natural language instructions (via...

A new technique for sending hidden messages is mathematically proven to escape detection

[ICLR 22] Latent Image Animator: Learning to Animate Images via Latent Space Navigation - GitHub - wyhsirius/LIA: [ICLR 22] Latent Image Animator: Learning to Animate Images via Latent Space Naviga...

A new neural network approach from MIT and IBM researchers captures the characteristics of a physical system's dynamic motion from video, regardless of rendering configuration or image differences, bringing us one step closer to creating realistic simulations from real-world videos.

An MIT machine-learning model proposes new molecules for drug discovery, making sure the suggested molecules can actually be synthesized in a laboratory.

MIT and IBM researchers develop an efficient machine learning method that uses chemical knowledge to create a learnable grammar with production rules to build new, synthesizable monomers and polymers.