MakinaRocks Research Published by Deep Learning Sector’s Premier Conference, ICLR 2020
The International Conference on Learning Representations (ICLR) selected MakinaRocks’s RaPP-based anomaly detection research to be published and presented in their eighth conference in 2020
MakinaRocks, a startup specializing in industrial AI solutions, announced on October 13th, the success of their RaPP-based anomaly detection research paper, selected and published by the International Conference on Learning Representations (ICLR), a prestigious global AI conference and organization renowned for their authority on the topic of deep learning.
The ICLR is known for promoting research in the field of artificial intelligence—gathering, presenting, and publishing papers related to deep learning for eight consecutive years. Due to the growing interest in artificial intelligence, approximately 1,600 papers are submitted each year. In 2020, MakinaRocks’s submission, a paper on their anomaly detection solution, “RaPP (Novelty Detection with Reconstruction along Projection Pathway),” was chosen to be presented and published at the ICLR’s eighth conference.
Recommended AI News: Extend and Oliver Join the Wells Fargo Startup Accelerator
RaPP, developed by MakinaRocks, uses deep neural networks to generate an anomaly score, improving the performance of anomaly detection. While standard anomaly detection models are limited to comparing the difference between the input and the output without consideration to the hidden values within the autoencoder, RaPP extends the process by leveraging hidden reconstruction errors produced by encoder.
Recommended AI News: BlackBerry QNX Adds Another Certification to its Embedded Software Portfolio
Through extensive performance evaluation with diverse image and sensor datasets, the model has proven to exceed other deep learning-based anomaly detection models in terms of performance. The Anomaly Detection Suite (ADS), developed by MakinaRocks, improves production efficiency by increasing failure prediction accuracy of critical machinery, thereby minimizing downtime.
Ki Hyun Kim, a principal machine learning research engineer at MakinaRocks, clarified, “Unlike previous anomaly detection models that propose new learning schemes or model architecture, RaPP is a relatively simple process that doesn’t require changing widely used autoencoders and makes use of currently existing models.”