Neural Networks - The Batch

Flowchart of Text-to-LoRA model processes task embeddings into LoRA adapters, showing weights and losses.

Machine Learning Research

LoRA Adapters On Tap: Text-to-LoRA generates task-specific LoRA adapters directly from natural language descriptions

The approach known as LoRA streamlines fine-tuning by training a small adapter that modifies a pretrained model’s weights at inference. Researchers built a model that generates such adapters directly.

Image illustrates data flow from raw satellite sources through processing to embeddings for climate tracking.

Machine Learning Research

Earth Modeled in 10-Meter Squares: Google’s AlphaEarth Foundations tracks the whole planet’s climate, land use, potential for disasters, in detail and at scale

Researchers built a model that integrates satellite imagery and other sensor readings across the entire surface of the Earth to reveal patterns of climate, land use, and other features.

Table comparing DINO, DINOv2, DINOv3, SigLIP 2, and PE on segmentation, depth estimation, tracking, and classification tasks.

Machine Learning Research

Better Image Processing Through Self-Supervised Learning: Meta’s DINOv3 gets an updated loss term and improved vision performance

DINOv2 showed that a vision transformer pretrained on unlabeled images could produce embeddings that are useful for a wide variety of tasks. Now it has been updated to improve the performance of its embeddings in segmentation and other vision tasks.

BitNet b1.58 matrix multiplication shows ternary weights enabling faster neural network computation.

Machine Learning Research

Low Precision, High Performance: Researchers at Microsoft and Tsinghua researchers propose 1.58-bit AI model that rivals full-precision competitors

Reducing the number of bits used to represent each parameter in a neural network from, say, 16 bits to 8 bits shrinks the network’s size and boosts its speed. Researchers took this approach to an extreme: They built a competitive large language model whose weights are limited to three values.

Dual line graphs showing factual QA accuracy and NLL against memory size for NQ and TQA datasets in AI models.

Machine Learning Research

Memory Layers for More-Factual Output: Meta researchers build Llama-style models that recall details without needing more computing resources

Improving a large language model’s factual accuracy typically requires making it bigger, which in turn, involves more computation. Researchers devised an architecture that enables models to recall relevant details without significantly increasing the amount of computation required.

A participant types while an MEG scan decodes brain activity into text in real-time, showing typed vs. decoded text.

Science

Reading Minds, No Brain Implant Required: Brain2Qwerty, a system that decodes thoughts using brain waves without surgery

To date, efforts to decode what people are thinking from their brain waves often relied on electrodes implanted in the cortex. New work used devices outside the head to pick up brain signals that enabled an AI system, as a subject typed, to accurately guess what they were typing.

Robotic arms collaborating to fold a red garment on a table.

Hardware

Household Help: π0, a machine learning system for household robotics

A new generation of robots can handle some household chores with unusual skill.

Deja Vu, an algorithm that accelerates inferencing of large language models

Machine Learning Research

Streamlined Inference: Deja Vu, a method that boosts LLM speed by activating only essential neural parts

It’s not necessary to activate all parts of a large language model to process a given input. Using only the necessary parts saves processing.

Science

Cross-Species Cell Embeddings: AI enhances cell type discovery, identifies previously elusive “Norn cells”

Researchers used an AI system to identify animal cell types from gene sequences, including a cell type that conventional approaches had discovered only in the past year.

Machine Learning Research

Better, Faster Network Pruning: Researchers devise pruning method that boosts AI speed

Pruning weights from a neural network makes it smaller and faster, but it can take a lot of computation to choose weights that can be removed without degrading the network’s performance.

Science

Early Detection for Pancreatic Cancer: A neural network shows remarkable accuracy in forecasting risk of pancreatic cancer.

A neural network detected early signs of pancreatic cancer more effectively than doctors who used the usual risk-assessment criteria. Researchers at MIT and oncologists at Beth Israel Medical Center in Boston...

SingSong's process for manufacturing instrumental music to accompany input vocals.

Machine Learning Research

Sing a Tune, Generate an Accompaniment: SingSong, a tool that generates instrumental music for unaccompanied input vocals

A neural network makes music for unaccompanied vocal tracks. Chris Donahue, Antoine Caillon, Adam Roberts, and colleagues at Google proposed SingSong, a system that generates musical accompaniments for sung melodies. You can listen to its output here.