Since the advent of LLM's like GPT4 everyone in various industries has been trying to harness their power. Healthcare is an industry where this is a specifically challenging problem due to the high accuracy requirements. Prompt Engineering is a common technique used to design instructions for model responses, however, its challenges lie in the fact that the generic models may not be trained to accurately execute these specific tasks. We will present our journey of developing a cost-effective medical LLM, surpassing GPT4 in medical note-writing tasks. We'll touch upon our trials with medical prompt engineering, GPT4's limitations, and training an optimized LLM for specific medical tasks. We'll showcase multiple comparisons on model sizes, training data, and pipeline designs that enabled us to outperform GPT4 with smaller models, maintaining precision, reducing biases, preventing hallucinations, and enhancing note-writing style.

Large-scale knowledge graphs (KGs) play a critical role in question answering over KGs (KGs-QA). Despite of large scale, KGs suffer from incompleteness which results in the low-precision of KGs-QA. Since existing researches of relation prediction process each triple independently, the hidden relations which are inherently present in the neighborhood can not be captured. Complementarily, to capture both entity and relation features in a given entity's neighborhood, an entity importance estimation network of attention-based graph embedding is proposed, which consists of the attention-based graph embedding module and the entity importance estimation module. The new embedding of an entity from its n-hop neighbor is learned by an attention-based graph embedding module, and then the learned embedding is integrated into the entity importance estimation module to find entities of high importance in n-hop neighbors of the central entity. Multi-hop relations are encapsulated and an auxiliary edge of n-hop neighbors is introduced, which realizes the relation prediction task. In addition, we are the first to realize KGs-QA while realizing relation prediction, which alleviates the phenomenon of missing relations and the low-precision problem of KGs-QA. As expected, KGs-QA's F1 increases significantly when relation prediction is invoked. In 80% missing relation, we observe a F1 increase of 21, 17, 18 and 16 points over the no-relation prediction configuration for SQ, WQ, GQ and QALD-5, respectively.

Service composition has been proven to be a convincing computing paradigm for rapidly constructing large-scale distributed applications within and across organizational boundaries. Quality of service (QoS)-aware service composition, i.e. selection of the optimal execution plan that maximizes the composition's end-to-end QoS properties, is an active area of research and development endeavors in service composition. In this article, we propose an orthogonal genetic algorithm (OGA) for QoS-aware service composition problem. Its significant feature is to incorporate an orthogonal design method into the initial population generation process and crossover operation. As a result, our algorithm is more robust and can search the solution space in a statistically sound manner. We have executed the OGA to solve 81 randomly generated service composition problems with different sizes and structures based on QWS data set including 2507 real Web services. The results indicate that our OGA can find near-optimal solutions within moderate numbers of generation and has the performance superiority in comparison with many existing optimization algorithms.