Probabilistic Numeric CNNs with Roberto Bondesan - #482
Published:May 10, 2021 17:36
•1 min read
•Practical AI
Analysis
This article summarizes an episode of the "Practical AI" podcast featuring Roberto Bondesan, an AI researcher from Qualcomm. The discussion centers around Bondesan's paper on Probabilistic Numeric Convolutional Neural Networks, which utilizes Gaussian processes to represent features and quantify discretization error. The conversation also touches upon other research presented by the Qualcomm team at ICLR 2021, including Adaptive Neural Compression and Gauge Equivariant Mesh CNNs. Furthermore, the episode briefly explores quantum deep learning and the future of combinatorial optimization research. The article provides a concise overview of the topics discussed, highlighting the key areas of Bondesan's research and the broader interests of his team.
Key Takeaways
- •The episode discusses Probabilistic Numeric Convolutional Neural Networks, which use Gaussian processes.
- •Other research from Qualcomm presented at ICLR 2021, including Adaptive Neural Compression and Gauge Equivariant Mesh CNNs, is mentioned.
- •The future of quantum deep learning and combinatorial optimization is briefly explored.
Reference
“The article doesn't contain a direct quote.”