深度學習硬體加速器設計
<br>
<br>Time: 3:30-6:20 pm, Thursday (R7R8R9)
<br>Place: Room 107, Delta Building
<br>Instructors:
<br>Youn-Long Lin (Two thirds)
<br>Albert Liu (One third), CEO Kneron 
<br>
<br>Intended Students: 
<br>CS Graduate Students or Undergraduate students in their 3rd, or 4th year 
<br>
<br>ob<x>jectives: 
<br>To teach students theory, algorithms, Python programming, Verilog 
<br>implementation, and ASIC optimization of contemporary neural networks in terms 
<br>of performance, accuracy, model size, energy efficiency
<br>
<br>Prerequisites:
<br>Digital Logic Design and Hardware Lab
<br>Computer Architecture
<br>Familiar with Synthesizable Verilog 
<br>
<br>Reference: 
<br>Selected papers
<br>Handouts
<br>
<br>Grading: 
<br>Machine problems
<br>Paper presentation
<br>Term Project
<br>Final Presentation
<br>
<br>Week	Date		Topics	Instructor	
<br>1	2/21		Multi la<x>yer Perceptron	Lin	
<br>2	2/28	Holiday	No meeting		
<br>3	3/7		Convolutional Neural Networks	Lin	
<br>4	3/14		Famous Networks: LeNet, AlexNet, ResNet, DenseNet	
<br>Lin	
<br>5	3/21		HW Accelerator – Eyerise, Cambricon	Lin	
<br>6	3/28		Model Compression – Hang Song’s Work	Lin	
<br>7	4/4	Holiday	No meeting		
<br>8	4/11				
<br>9	4/18		Memory Computing and fast IO	Chunchen	
<br>10	4/25		Reconfigurable Neural Network 
<br>-Convolution Filter Decomposition Technique (la<x>yer Level Reconfiguration)
<br>	Chunchen	
<br>11	5/2				
<br>12	5/9				
<br>13	5/16		Accelerators for Cloud Training - Example 
<br>accelerators: - TPU (Google's Tensor Processing Unit), Wave computing 	
<br>Chunchen	
<br>14	5/23		Convolutional Neural Network operation device 	
<br>Chunchen	
<br>15	5/30				
<br>16	6/7		Accelerators for Inference: - Example accelerators: - 
<br>Kneron, Minerva	Chunchen	
<br>17	6/14				
<br>18		Final Exam Week			
<br>