Nehodí sa? Žiadny problém! U nás môžete do 30 dní vrátiť
S darčekovým poukazom nešliapnete vedľa. Obdarovaný si za darčekový poukaz môže vybrať čokoľvek z našej ponuky.
30 dní na vrátenie tovaru
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.