Buch, Englisch, 172 Seiten, Format (B × H): 148 mm x 210 mm, Gewicht: 258 g
Reihe: Schriftenreihe / Institut für Mess- und Regelungstechnik, Karlsruher Institut für Technologie
Buch, Englisch, 172 Seiten, Format (B × H): 148 mm x 210 mm, Gewicht: 258 g
Reihe: Schriftenreihe / Institut für Mess- und Regelungstechnik, Karlsruher Institut für Technologie
ISBN: 978-3-7315-1474-9
Verlag: Karlsruher Institut für Technologie
We address interpretable representation learning for motion forecasting in self-driving cars. Rather than treating transformers as black boxes, we develop methods to interpret and modify learned representations. We introduce self-supervised pre-training with interpretable objectives. Moreover, we probe latent spaces of forecasting models and reveal interpretable features, allowing us to make targeted interventions. Finally, we uncover retrocausal mechanisms, which enable goal-based instructions.




