Question answering is a natural language understanding task that involves reasoning over both explicit context and unstated, relevant domain knowledge. Large language models (LLMs), which underpin most contemporary question answering systems, struggle to induce how concepts relate in specialized domains such as medicine. Existing medical LLMs are also costly to train. In this work, we present MEG, a parameter-efficient approach for medical knowledge-augmented LLMs. MEG uses a lightweight mapping network to integrate graph embeddings into the LLM, enabling it to leverage external knowledge in a cost-effective way. We evaluate our method on four popular medical multiple-choice datasets and show that LLMs greatly benefit from the factual grounding provided by knowledge graph embeddings. MEG attains an average of +10.2% accuracy over the Mistral-Instruct baseline, and +6.7% over specialized models like BioMistral. We also show results based on Llama-3. Finally, we show that MEG's performance remains robust to the choice of graph encoder.

Furthermore, while the neuromorphic community has argued in favor of their higher We seek to enable classic processing of continuous energy efficiency for decades, recent research and breakthroughs ultra-sparse spatiotemporal data generated by in edge AI accelerators suggest that this might not event-based sensors with dense machine learning be the case (Dampfhoffer et al., 2023; Garrett et al., 2023; models. We propose a novel hybrid pipeline composed Moosmann et al., 2023; Caccavella et al., 2023). of asynchronous sensing and synchronous processing that combines several ideas: (1) an embedding Nevertheless, considering the inherent advantages of eventbased based on PointNet models - the ALERT vision sensors, namely high dynamic range (HDR) module - that can continuously integrate new and and high temporal resolution - simultaneously, without any dismiss old events thanks to a leakage mechanism, tradeoffs between the two -, we aim to find a way to leverage (2) a flexible readout of the embedded data this sparse and low-latency data for real-world situations.