Beyond the Signals: Multimodal Learning for Seizure Control
A new wave of research is combining data from brain scans, wearable sensors, and even video to improve the accuracy and speed of epileptic seizure detection and prediction.
Science
Catching Flare Ups: A New View of Solar Activity
Researchers are improving solar flare prediction by integrating far-side observations with surface magnetic field modeling, offering a more comprehensive view of the sun’s volatile behavior.
Distorting the Signal: How Skeletonization Fools Vision-Language Models
A new attack method subtly alters mathematical formulas to mislead even the most advanced AI systems like ChatGPT during text recognition.
Speeding Up the Cosmos: AI for Astrophysical Simulations
New research demonstrates how machine learning techniques are dramatically accelerating radiative hydrodynamics simulations, offering a path to more detailed understanding of complex astrophysical phenomena.
AI Guardians for the Grid: Detecting Attacks on Power System Relays
Researchers are exploring the use of artificial intelligence to bolster the security of critical power infrastructure against increasingly sophisticated cyber threats.
Building Trust in the Intelligent Factory
A new framework aims to bridge the gap between data, services, and knowledge, enabling truly resilient and trustworthy industrial intelligence for the era of Industry 5.0.
Rewarding Collaboration: A New Approach to Federated Learning
![Client participation in Federated Learning exhibits network effects, wherein increased client involvement demonstrably amplifies the overall model accuracy, a phenomenon quantified by the observation that the marginal gain in performance diminishes as participation nears saturation, mirroring a logarithmic relationship described by [latex] \log(N) [/latex], where <i>N</i> represents the number of participating clients.](https://arxiv.org/html/2601.04648v1/x1.png)
This research introduces a mechanism to incentivize participation in federated learning systems, addressing the complex interplay of network effects and application-specific needs.
Decoding the Language Brain: How Multilingual Models Specialize
![The study demonstrates that transferring neuron masks-specifically, those identified within a [latex]LLaMA2-7B-Base[/latex] model-directly to its [latex]LLaMA2-7B-Chat[/latex] counterpart yields measurable performance on target-language benchmarks-MMLU, C-Eval, and Belebele\_vi-suggesting a shared underlying neural infrastructure despite differing conversational objectives.](https://arxiv.org/html/2601.04664v1/x4.png)
New research reveals how individual neurons within large language models become dedicated to specific languages, moving beyond simple correlation to prove functional necessity.
Beyond Confidence: Smarter Alert Triage with Decision-Aware Trust
![Cost-weighted loss is shown to vary with the escalation threshold when using calibrated probabilities, with the decision-aware threshold of 0.0909 derived from an asymmetric cost structure where the cost of a false negative ([latex]C_{FN}=10[/latex]) is ten times greater than the cost of a false positive ([latex]C_{FP}=1[/latex]).](https://arxiv.org/html/2601.04486v1/cost_vs_threshold_rf.png)
New research demonstrates that calibrating trust signals to the real-world costs of security decisions dramatically improves the efficiency and accuracy of Security Operations Centers.
Building Robust Teams of AI Agents
New research details a framework for optimizing the structure and communication of AI agent groups to withstand failures and maximize performance.