As humans, we love the uncertainty that comes with predictions. For example, we always want to know what the chances are of it raining before we leave the house. However, with traditional deep learning, we only have a point prediction and no notion of uncertainty. Predictions from these networks are assumed to be accurate, which is not always the case. Ideally, we would like to know the level of confidence of predictions from neural networks before making a decision.
For example, having uncertainty in the model could have potentially avoided the following disastrous consequences:
- In May 2016, the Tesla Model S crashed in northern Florida into a truck that was turning left in front of it. According to the official Tesla blog (https://www.tesla.com/en_GB/blog/tragic-loss), Neither Autopilot nor...
