Obálka

Interpretable Representation Learning for Motion Forecasting

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 pr...

Celý popis

Uloženo v:
Hlavní autor: Wagner, Royden
Médium: Online
Jazyk:angličtina
Vydáno: KIT Scientific Publishing 2026
Témata:
Autonomes Fahren
Bewegungsvorhersage
Maschinelles Lernen
Mechanistic interpretability
Motion forecasting
Representation learning
Self-driving cars
Transformer models
thema EDItEUR::U Computing and Information Technology::UY Computer science
On-line přístup:ONIX_20260519T105721_9783731514749_14
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Shrnutí: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.

Podobné jednotky