#computer-society #machine-learning #computer-science #neural-networks #learning-models

Popular Large Language Models are Machine Learning (ML) models first trained on a large Copus of textual general context subsequently fine-tuned to be user friendly task specific. The performance of all machine learning models including Deep Neural Networks (DNNs) depend on the context available to train them. This context may be consisting of two forms: specific context from sample-level experimental data, and general context from knowledge accumulated and curated over the human history. Current DNNs rarely consider this general context.  In our experience, neither form of context is sufficient alone: experimental data are sensitive to dataset specific noise and lack broader contextual grounding, while curated knowledge may lack the resolution required for the specific problem. Consequently, most DNNs relying solely on experimental data risk learning spurious correlations rather than underlying context. Here we introduce Knowledge Inclusive Machine Learning (KIML), a paradigm that integrates both context types within a unified analytical pipeline. KIML combines experimental data with various types of general context: existing mathematical models (equations), literature-derived representations from research paper data bases and structured knowledge graphs etc.