Beyond Standard Deep Learning Models for Time Series and Sequences

Explore advanced techniques for improving deep learning models in time series and sequence analysis in this 32-minute conference talk by Professor Diego Klabjan from Northwestern University. Delve into novel concepts that enhance the performance of recurrent neural networks and transformers for temporal data. Learn about dynamic confident prediction output and adaptive computational time models that address challenges in time series data. Discover how these models dynamically allocate layers and computational resources based on data complexity. Examine the application of sequence-to-sequence models with attention for sparse temporal features. Gain insights from real-world examples using both proprietary and public financial instrument datasets, demonstrating the practical implications of these advanced techniques in the field of machine learning for time-based data.