Research author
Armando J. Pinho
3 AI research papers in the One9Founders library, with summaries and links to original sources.
Papers by Armando J. Pinho
Fast and Interpretable Autoregressive Estimation with Neural Network Backpropagation
Anaísa Lucena, Ana Martins, Armando J. Pinho, et al.
This paper proposes a new method for estimating parameters in autoregressive (AR) time series models by using neural networks instead of traditional statistical methods. The key advantage is that the neural network approach is much faster (up to 34x speedup), more reliable (doesn't fail like conventional methods do 55% of the time), and still produces accurate, interpretable results comparable to traditional approaches.
Revisiting OmniAnomaly for Anomaly Detection: performance metrics and comparison with PCA-based models
Bruna Alves, Ana Martins, Armando J. Pinho, et al.
This paper compares OmniAnomaly, a popular deep learning model for detecting anomalies in time series data from multiple sources, against a simpler traditional statistical method called PCA. Using the same fair evaluation standards across 100 runs, the researchers found that the simpler PCA method performs just as well or even better than the complex deep learning model, suggesting that how we measure and compare these models matters more than we thought.
Unified Taxonomy for Multivariate Time Series Anomaly Detection using Deep Learning
Bruna Alves, Armando J. Pinho, Sónia Gouveia
This paper presents a comprehensive classification system (taxonomy) for organizing deep learning methods used to detect anomalies in multivariate time series data. The researchers identified eleven key dimensions across three categories (Input, Output, and Model) by analyzing existing research and found that the field is increasingly moving toward Transformer-based models and approaches that either reconstruct or predict data patterns.