output=T1（input1）T2（input2）+T3（input3）

why input3 is needed here?

and why it is accumulated in matrix multiplication?

output=T1（input1）T2（input2）+T3（input3）

why input3 is needed here?

and why it is accumulated in matrix multiplication?

You should probably websearch “convolution neural networks matrix multiply accumulate” and do some reading.

Thanks for reply.

The Multiply kernel is more like fully connect in CNN