OOD
A situation in which the test data distribution differs significantly from the training data distribution. Deep reinforcement learning policies have limited extrapolation capability when faced with OOD inputs and tend to output high-confidence erroneous predictions, making this a key dimension for measuring the safety of navigation policies.
Timeline (last 90 days)
Deep reinforcement learning policies perform well in distribution, but neural networks have limited extrapolation capability when faced with OOD inputs, and easily output high-confidence yet erroneous predictions, leading to dangerous decisions.
OOD samples may fall outside the coverage, thereby exhibiting high uncertainty.
The neural expansion framework explains OOD samples as inputs that fall outside the ray coverage of training samples, so the model's predictions on them lack stability and may exhibit high uncertainty or random predictions.
Linear probes perform well on the training distribution, but their performance often drops sharply when facing out-of-distribution (OOD) samples.
All Facts (4)
Deep reinforcement learning policies perform well in distribution, but neural networks have limited extrapolation capability when faced with OOD inputs, and easily output high-confidence yet erroneous predictions, leading to dangerous decisions.
50%UnverifiedOOD samples may fall outside the coverage, thereby exhibiting high uncertainty.
50%UnverifiedThe neural expansion framework explains OOD samples as inputs that fall outside the ray coverage of training samples, so the model's predictions on them lack stability and may exhibit high uncertainty or random predictions.
50%UnverifiedLinear probes perform well on the training distribution, but their performance often drops sharply when facing out-of-distribution (OOD) samples.
50%