Our Team Innovation Directions
Conformal Prediction
We have been working to improve Conformal Prediction, an AI “margin of error” that often yields overly broad answers just to stay safe. Our method fixes this by adapting the AI’s precision—providing highly specific answers for straightforward data and wider ranges for complex situations.
Application Scope(s): General
Innovations:
- 2026/Jun [ILLIANCE]: Preprint made available 10.2139/ssrn.6880624
- 2026/Feb [ILLIANCE]: Creation of an holistic measure for classification settings
- 2026/Jan [ILLIANCE]: Combine local data complexity with non-conformity scores
Forecasting of Zero-Inflated Time Series
We have being working to develop the BEST algorithms for forecasting of zero-inflated time series. This series are quite common in ral problems when there a on-off situation (e.g. sell value/no sell or capacity/empty).
Application Scope(s): Logistics, Manufacturing
Innovations:
- 2026/Fev [ILLIANCE]: (to be added)
- 2025/Mar [NEXUS]: (to be added)
Tree-based Machine Learning Models
We have being working to develop algorithms that can better deal with DIFFICULT real-data.
Application Scope(s): General
Innovations:
- 2026/Mar [ILLIANCE]: Preprint made available for the improvements on streaming 10.2139/ssrn.6442025
- 2025/Dec [ILLIANCE]: Publication of the improved batch algorithms 10.7717/peerj-cs.3319
- 2025/Oct [ILLIANCE]: Adaptation of the new criteria for streaming
- 2025/Jun [ILLIANCE]: Adaptation of the new criteria for Tree-based batch models
- 2025/May [ILLIANCE]: Creation of a split criteria based on the Ultra Generalized Entropy
- 2024/Jun [CIDMA]: New mathematical concept of Ultra-Generalization of Entropy
Weak/Strong Time Series Anomalies
We have being working to develop algorithms that UNCOVER anomalies in (quasi)periodicity, which are undetected by traditional statistical techniques or machine learning techniques.
Application Scope(s): General
Innovations:
- 2025/Jun [ILLIANCE]: New relevant anomaly detection algorithm by Algebraic Topology techniques