Yehudit Aperstein

Synthetic clinical communication can effectively bootstrap NLP systems, outperforming traditional zero-shot approaches and paving the way for more robust healthcare applications.

Integrating wavelet-derived features into UAV monitoring boosts rip-current detection accuracy to over 95%, revealing critical insights for beach safety.

Anomaly detection in cyber-physical systems can be dramatically improved by modeling normal behavior as a complex, multimodal distribution rather than a simplistic blob.

Students' ability to effectively engage with generative AI hinges on mastering distinct skills, not just prompting.

Generative models stumble when extracting clinical follow-up instructions, but a hybrid neural-symbolic approach nails it, achieving near-perfect F1 and date accuracy even on unseen actions.

After 15 years without a new HMX-class energetic material, AI has delivered a DFT-validated compound with a detonation velocity of 9.00 km/s, ready for synthesis after just a few GPU-days.

Synthetic data, when carefully crafted, can provide a surprisingly effective benchmark for educational sentiment analysis, even rivaling the performance of large language models in zero-shot settings.

LLMs can be steered to simulate students with specific strengths and weaknesses, opening the door to AI-driven teacher training.

Robots can collaboratively allocate tasks with near-optimal performance even with *zero* communication, prior knowledge, or coordination, by exploiting low-rank structure in task suitability.