1. Within the field of information retrieval, Target Retrieval (often referred to as Detection-Based Retrieval in remote sensing) has remained a vibrant and central research topic for several decad...
Keyframe extraction aims to sum up a video's semantics with the minimum number of its frames. This paper puts forward a Large Model based Sequential Keyframe Extraction for video summarization, dubbed LMSKE, which contains three stages as below. First, we use the large model "TransNetV21" to cut the video into consecutive shots, and employ the large model "CLIP2" to generate each frame's visual feature within each shot; Second, we develop an adaptive clustering algorithm to yield candidate keyframes for each shot, with each candidate keyframe locating nearest to a cluster center; Third, we further reduce the above candidate keyframes via redundancy elimination within each shot, and finally concatenate them in accordance with the sequence of shots as the final sequential keyframes. To evaluate LMSKE, we curate a benchmark dataset and conduct rich experiments, whose results exhibit that LMSKE performs much better than quite a few SOTA competitors with average F1 of 0.5311, average fidelity of 0.8141, and average compression ratio of 0.9922.
Virtual-patient models need a way to represent a person’s biological state before they can model how that state may change. Many AI-for-science models focus on molecules, cells, or a single organ, while treatment decisions are made for people who may have several conditions and many possible interventions. We used drug repurposing as a human-scale test of whether organized biological modules can help rank known drug–disease relationships.We introduce SteeraMed Bench, which compares panels drawn from a 332-module library across drug-repurposing tasks. The benchmark included 1,916 small molecules and five primary chronic-disease tasks, with an exploratory analysis across 23 disease categories. In standard cross-validation, recall@20—the fraction of known positive drug–disease pairs found among the top 20 ranked drugs—was 0.494 for the 117-module nutraceutical panel across five diseases, close to 0.524 for the full library. In the 23-category analysis, the full library ranked first in only 9 categories; the best-performing panel varied by disease: aging-hallmark panels led for type 2 diabetes and osteoporosis, food-as-medicine for depression, and nutraceutical modules for atherosclerosis/hyperlipidemia. These findings show that different biological panels carry different ranking signals within the benchmark. Performance fell when drugs with related target families were kept out of training, and was near chance when an entire disease was held out.We also tested whether a language-model-assisted agent could propose new modules for the library. Two of 14 proposals showed nominal positive gains in the second round, but neither remained significant after correction for testing seven concepts. SteeraMed Bench provides a shared way to compare and update biological representations for future virtual-patient research. It is a benchmark for representation and candidate prioritization, not a complete virtual patient or a validated drug-discovery system.