Zoom Logo

IEMCON 2020 DAY 4 KEYNOTE - Shared screen with speaker view
Son Vuong (UBC) - IEMCON General CoChair
01:23:41
Please, feel free to type in your questions as you have them, that will be answered by the speaker during QA period after the presentation
nikesh Singh
01:24:58
What are the operations of concatenation or combination exist in the literature for CNN architecture of Deep Learning ?
Partha Bakshi
01:43:28
very very nice presentation. My question is - What Is the Difference Between a Feedforward Neural Network and Recurrent Neural Network? How it is related to CNN architecture?
nikesh Singh
01:46:45
Is there any algorithm that helps to distinguish between vowels and consonants in the waveform signal?
Partha Bakshi
01:53:30
The difference between consonants and vowels is linguistical. Eg. 'm', 'n', 'l' sounds are quite similar to vowels, but they don't form syllable. A speech recognizer can provide with segmentation data as well. As sounds are different from waves so I guess a particular algorithm can't distinguish between vowels & consonants
Satyasaran Changdar
01:56:15
How to decide the number of hidden layers and nodes in a hidden layer in DNN?
Gerald Penn
02:00:51
Nikesh, I'm sorry, I don't understand your question about concatenation and combination. Do you mean concat/combo of neural layers or of text or speech input?
Gerald Penn
02:02:33
@Satyasaran: deciding the number of hidden layers and nodes in a DNN is often determined experimentally, by tuning using the development test set data that I referred to at the beginning of the talk
Satyasaran Changdar
02:04:04
Okay thank you very much sir.
Gerald Penn
02:04:11
There are some basic rules of thumb that are emerging. For example, the model size should probably scale up like the square of the dataset size
Gerald Penn
02:04:23
This is a result recently published by Jimmy Ba
Gerald Penn
02:04:47
But what you have to understand about those rules is that their purpose is to avoid overfitting
Gerald Penn
02:05:37
The actual number of nodes or layers still needs to be experimentally determined, guided by a proportion like that, because it is so domain/problem-dependent
Satyasaran Changdar
02:07:12
Thank you sir.
Son Vuong (UBC) - IEMCON General CoChair
02:32:39
Please, feel free to type in your questions as you have them, that will be answered by the 2nd speaker during QA period after the presentation
nikesh Singh
02:36:06
How can we identify the points that are poking out from a point cloud plot?
nikesh Singh
02:37:24
How to set height values of point cloud instead of gray values of an image?
SmartSociety Conferences
02:41:48
SESSION 25: https://zoom.us/j/95678695234?pwd=OUlldnd4bHhON01EWGpDLzJKbFBFQT09SESSION 26: https://zoom.us/j/97835877906?pwd=azN3ZTZmRHUydzZzWXBqNTlkTmZudz09SESSION 27: https://zoom.us/j/97395180621?pwd=b2lPMTliYnFvOTZyRmpSNmtrRCtxQT09SESSION 28: https://zoom.us/j/98227487536?pwd=S24zdEdLc2k1UW5JbzZUQ21RYTVpQT09Valedictory Session: https://zoom.us/j/94280259832?pwd=SHJzN0JyeDhzTmxsbklOZUJjNzhaZz09
Edson Ursini
02:43:32
Thank you!