🔬 AI Research

Academic papers, research breakthroughs, and scientific advances in machine learning and AI.

585 articles

Nested Learning: a theory that could reshape the future of AI

For more than a decade, AI progress has mostly meant building ever-larger neural networks with more layers and parameters. An essay by Raffaella Aghemo explores nested learning as a theory that could challenge this scale-first path. Source: medium.datadriveninvestor.com

SURGE teaches graph AI to forget by treating deletions as structural changes

Researchers have introduced SURGE, a graph unlearning framework that handles data deletion requests as structural perturbations to the graph. It then repairs the affected graph neural network so the removed data no longer influences the model. Source: bioengineer.org

$12.8M NIH grant funds ML and AI push to bring genomics into the clinic

Scientists know far more pathogenic gene variants than there are clinical options for treating them. A new NIH-funded project led by Vanderbilt will use machine learning and AI to help close that gap. Source: news.vumc.org

Researchers chart how AI hardware has evolved over the years

Since 2018, the Lincoln Laboratory Supercomputing Center has kept a running survey of the newest AI hardware. The aim is to help researchers and the U.S. government keep track of the fast-moving field. Source: news.mit.edu

Study: ML models can forecast suicide risk among U.S. Army soldiers

A study finds that machine-learning models can estimate the risk of suicide and nonfatal suicide attempts among regular U.S. Army soldiers. The analysis covered more than 580,000 soldier records. Source: usmedicine.com

Study of ~10,000 papers finds AI biomimetics booming, but energy lags behind

A bibliometric analysis of nearly 10,000 publications shows AI-driven biomimetics thriving in robotics and computing. Energy applications, however, remain notably underrepresented. Source: bioengineer.org

Explainable AI opens the black box of crash severity prediction

A systematic review of 237 studies from 2014 to 2024 shows how explainable AI methods such as SHAP and LIME are making machine learning models for crash severity prediction more transparent and interpretable. Source: bioengineer.org

Cantina's open apex-flash-1 model solves 40 of 60 held-out security bug tasks

Cantina Security released apex-flash-1, an MIT-licensed 321B-parameter model built for security research. It reaches 66.7% pass@1 on held-out bug tasks while costing 31 times less than Opus. Source: marktechpost.com

Blockchain and ML team up to detect money laundering with 96% accuracy

Researchers paired blockchain-based customer authentication with an ensemble of three machine learning classifiers to spot money laundering. The combined system reaches roughly 96% detection accuracy. Source: bioengineer.org

Google brings federated learning into TEEs, giving Gboard verifiable differential privacy

Google Research moved federated learning into trusted execution environments (TEEs). Gboard now trains with externally verifiable central differential privacy, and server-side training runs faster. Source: marktechpost.com

ML models predict strength of cement-free green concrete

Researchers trained six machine learning models on 2,048 laboratory records to forecast the compressive strength of cement-free concrete. Gradient boosting came out as the best predictor, offering a data-driven route to designing greener construction materials. Source: bioengineer.org

Teaching Speech Models to Tell Languages Apart Boosts Multilingual Learning

Multilingual self-supervised speech models can gain from sharing information across languages, but the benefit is harder to realize when the total amount of pretraining data is held constant. The work looks at how explicitly discriminating between languages can improve what these models learn about linguistic structure. Source: machinelearning.apple.com

New Federated Learning System Offers Verifiable Privacy Guarantees

Researchers announced a Federated Learning system that provides externally verifiable privacy guarantees. It achieves this while shifting more of the computation to the server to improve efficiency. Source: research.google

Dual-Unitary Circuits Give Quantum Reservoir Computing a Boost

Using dual-unitary circuits in a brickwork architecture offers a new platform for quantum reservoir computing, a machine learning technique. Researchers report that this setup improves the approach's performance. Source: quantumzeitgeist.com

Tuned Quantum Machine Learning Helps Researchers Curb Overfitting

New research examines how a quantum kernel affects active learning when paired with Gaussian Process Regression (GPR). The framework's performance depends strongly on how the kernel is tuned. Source: quantumzeitgeist.com

RLTL;DR: Agents Learn by Absorbing Their Own Feedback

The standard approach in reinforcement learning with verifiable rewards (RLVR) lets agents make several attempts at a task and then optimizes based on the outcomes. This work explores self-improvement by internalizing feedback that the agent generates for itself. Source: machinelearning.apple.com

New closed-loop IoT engine fuses machine learning with operations research

IoT systems produce a constant flood of sensor data, but converting that raw stream into truly good decisions is still a major challenge. A new closed-loop decision engine combines machine learning with operations research to bridge that gap. Source: bioengineer.org

SwRI uses generative AI to raise the odds of drug development success

Southwest Research Institute (SwRI) is applying a generative AI tool to pharmaceutical development. The machine learning toolkit aims to improve the chances that drug candidates succeed. Source: medicalxpress.com

Quandela taps quantum fingerprints to make machine learning more data-efficient

Research from Quandela examines how photonic quantum computing can use quantum fingerprints to boost the data efficiency of machine learning. The approach is especially relevant when labeled data is scarce. Source: quantumzeitgeist.com

Study Probes How Language Models Lose Structure When Serializing Expressions

Researchers examine what happens when language models, during chain-of-thought reasoning or when passing free-text intermediates, convert structured information into natural language. Their round-trip study of tree-structured expressions tests how much of that structure survives serialization, exposing a communication bottleneck. Source: machinelearning.apple.com