Task: Demonstrate How Loss and Accuracy Can Increase Simultaneously in a Neural Network
Task: Demonstrate How Loss and Accuracy Can Increase Simultaneously in a Neural Network: a task in Terminal-Lego-15k (Harbor dataset). In deep learning, it is commonly assumed that loss and accuracy are always inversely related — when loss goes down, accuracy goes up, and vice versa. However, this…
The task
In deep learning, it is commonly assumed that loss and accuracy are always inversely related — when loss goes down, accuracy goes up, and vice versa. However, this is not always the case. There are scenarios where both loss and accuracy increase (or both decrease) simultaneously during training.
Part of PrimeIntellect/Terminal-Lego-15k.