Lecture, Conference Presentation, Keynote
Chris Re
- Expectation that ETL pipeline claims regarding quality surpassing human annotators will be validated mid-presentation
- Prediction that probabilistic inference will reduce development effort by orders of magnitude compared to traditional pipeline construction
- Plan to leverage modern hardware parallelism, including SIMD, multi-core, and NUMA architectures, to build scalable inference engines
- Anticipation of a structured database containing the world's largest fossil record derived from PDFs and charts to analyze historical die-offs
- Projection that scientific knowledge will remain uniquely accessible yet largely unreadable for specific narrow questions
- Goal to extend the proposed approach across diverse scientific domains, including drug repurposing and genomics
- Plan to expand human trafficking application pilots to additional law enforcement agencies within the next couple of months
- Strategy to implement the "dumbest algorithm" (Gibbs sampling) at maximum speed using a highly optimized engine
- Assertion that relaxing consistency for statistical algorithms is currently in its infancy but may evolve for future energy efficiency and performance
- Expectation of achieving speedups of 10, 100, or 1,000 times over competitor systems via hardware parallelism techniques like Hogwild