Brief summary: if you repeatedly flip an unfair coin that’s more likely to land on heads, the most likely result is all heads. However, this isn’t a typical result. To get a typical result you...
Brief summary: if you repeatedly flip an unfair coin that’s more likely to land on heads, the most likely result is all heads. However, this isn’t a typical result. To get a typical result you need to get tails sometimes.
Similarly, in machine learning aiming for the most likely result is likely to give you something weird and you want to aim for a typical result instead. In real life, low-probability events happen sometimes and you need to include some of them.
Brief summary: if you repeatedly flip an unfair coin that’s more likely to land on heads, the most likely result is all heads. However, this isn’t a typical result. To get a typical result you need to get tails sometimes.
Similarly, in machine learning aiming for the most likely result is likely to give you something weird and you want to aim for a typical result instead. In real life, low-probability events happen sometimes and you need to include some of them.