Research
Müntz-Szász Networks: Rethinking AI for Physics Problems
Müntz-Szász Networks (MSN) use learnable fractional power functions to boost neural network accuracy and efficiency in physics-informed tasks.
Forgetting Neural Networks: A New Frontier for Data Privacy in AI
Researchers introduce Forgetting Neural Networks, a brain-inspired method that lets AI models erase specific training data without losing overall performance.
New Framework Reveals Overstated Robustness in Spiking Neural Networks
Researchers introduce methods that expose weaknesses in Spiking Neural Networks’ defenses against adversarial attacks.
AgenticTCAD Cuts Semiconductor Design Time from Days to Hours
AgenticTCAD uses AI-driven automation to speed up device design, challenging traditional methods and shaking up the semiconductor industry.
Stanford AI Lab Finds Selective Classification Can Worsen Medical AI Bias
Selective classification boosts overall accuracy but may increase disparities in critical medical AI diagnoses, calling for cautious deployment.
AI Tutor LearnLM Matches Human Tutors in UK School Trial
LearnLM, a generative AI model, matches or outperforms human tutors in UK schools, pointing to a future of scalable, personalized education.
Schrödinger AI: Bringing Quantum Mechanics to Machine Learning
Schrödinger AI applies quantum physics principles to reshape AI training and reasoning.
New Metric STED Boosts Consistency in AI-Generated Structured Data
Researchers unveil STED, a metric that sharpens reliability in LLM outputs, with Claude-3.7-Sonnet leading the pack.
Sat-EnQ Cuts Variance, Boosts Stability in Deep Q-learning
Sat-EnQ uses a two-step satisficing method to make reinforcement learning more stable and efficient.
Partial-LoRA Cuts Parameters by 87% Without Losing Accuracy
New research shows Partial-LoRA trims model fine-tuning size drastically while matching or beating accuracy.
AI Foundation Model Spots Earthquake Damage From Space with New Precision
Using the SATLAS model, researchers detected subtle surface ruptures from the 2023 Turkey-Syria quake that traditional methods missed.
New Research Challenges Universal Models for Time Series Data
A recent paper argues that one-size-fits-all models for time series data fall short, urging a move to specialized, context-aware agents instead.
ASG-SI: A Framework to Secure and Govern Self-Improving AI
ASG-SI introduces an auditable skill graph to boost AI accountability, tackling reward hacking and behavioral drift.
Why Top AI Models Still Fail High School Geometry
GeoBench reveals that vision-language models like OpenAI-o3 aren’t reasoning through geometry—they’re just recalling answers.
PATHWAYS Benchmark Reveals Critical Reasoning Failures in Web-Based AI Agents
A new benchmark exposes how current web-based AI agents stumble on multi-step reasoning, often fabricating their decision process and falling prey to misleading information.