Timeline for How are biases updated when 'batch size' > 1?
Current License: CC BY-SA 4.0
14 events
when toggle format | what | by | license | comment | |
---|---|---|---|---|---|
Oct 23, 2018 at 14:08 | answer | added | François Darmon | timeline score: 1 | |
Oct 23, 2018 at 13:59 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
added 77 characters in body
|
Oct 23, 2018 at 13:45 | comment | added | Sebastian Nielsen | @user7573566 My loss function is MSE (mean squared error) $1/M∗(a^3−y)$ where M is the total number of training examples. I left $1/M$ in the above to simplify the example. What scalar function would you recommend I used instead, so the derivative is becomes $3\times1$ vector? | |
Oct 23, 2018 at 13:37 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
deleted 6 characters in body
|
Oct 23, 2018 at 13:05 | comment | added | Sebastian Nielsen | @user7573566 Would it be incorrect to simply average each row, so that the current $3\times 5$ matrix becomes a $3\times 1$ vector? (I refer to $\partial E/\partial b^3$) | |
Oct 23, 2018 at 9:27 | comment | added | François Darmon | What is you loss function ? It is supposed to be a scalar function so the derivative should be a $3 \times 1$ vector. Also, you never use the bias, are you sure that for example $a^1 = W^1 \cdot X $ instead of $a^1 = W^1 \cdot X + b^1 $ | |
Oct 22, 2018 at 20:34 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
deleted 150 characters in body
|
Oct 22, 2018 at 20:23 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
deleted 150 characters in body
|
Oct 22, 2018 at 17:51 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
added 146 characters in body
|
Oct 22, 2018 at 17:46 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
added 146 characters in body
|
Oct 22, 2018 at 16:27 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
deleted 2 characters in body
|
Oct 22, 2018 at 16:20 | history | edited | Sebastian Nielsen | CC BY-SA 4.0 |
deleted 2 characters in body
|
Oct 22, 2018 at 16:05 | review | First posts | |||
Oct 22, 2018 at 16:41 | |||||
Oct 22, 2018 at 16:03 | history | asked | Sebastian Nielsen | CC BY-SA 4.0 |