The AI Canvas Newsletter #10

Exponentially Faster Language Modelling – ETH Zurich

“Language models only really need to use an exponential fraction of their neurons for individual inferences. As proof, we present UltraFastBERT, a BERT variant that uses 0.3% of its neurons during inference while performing on par with similar BERT models.”

A guide to LLM inference and performance – Baseten

“We want to use the full power of our GPU during LLM inference. To do that, we need to know if our inference is compute bound or memory bound so that we can make optimizations in the right area. Calculating the operations per byte possible on a given GPU and comparing it to the arithmetic intensity of our model’s attention layers reveals where the bottleneck is: compute or memory. We can use this information to pick the appropriate GPU for model inference and, if our use case allows, use techniques like batching to better utilize our GPU resources.”

Large-scale pancreatic cancer detection via non-contrast CT and deep learning

A recent study in Nature Medicine unveils a deep learning model, PANDA, that significantly improves the detection of pancreatic ductal adenocarcinoma (PDAC) using non-contrast CT scans. The model outperforms average radiologist sensitivity by over 30% and showcases a potential for large-scale, non-invasive cancer screening with high specificity, offering a promising tool for early detection in both symptomatic and asymptomatic individuals.

RAGs

“Build ChatGPT over your data, all with natural language.”

ML-Papers-Explained

“Explanation to key concepts in ML

Dive into 'The AI Canvas', our podcast exploring the transformative potential of generative AI. Engage in fireside chats, case studies, and innovative discussions on AI’s impact on industries and creativity.