5. AI: Artificial intelligence is a branch of science which is into
making machines think like humans.
AI/ML/DL the buzzwords…
DL: Deep Learning is a branch of machine learning that involves layering
algorithms in an effort to gain greater understanding of the data.
ML: Machine Learning is a collection of algorithms that
can learn from and make predictions based on recorded
data, optimize a given utility function under uncertainty,
extract hidden structures from data and classify data into
concise descriptions.
6. QuAI is QNAP’s AI Developer Package and is
intended for data scientists and developers to
quickly build, train, and optimize their AI Models on
a QNAP NAS.
Riding the AI wave with QNAP ‘QuAI’…
+ = QuAI
AI QNAP NAS
QNAP’s AI Developer
Package
7. Who is it for…
Data Scientists Engineers Students
8. Deep Learning Application Development
Process
Data Scientist is presented with
Business Goal
He puts together dataset and target to accomplish that
business goal
He goes through iterative
process to build and optimize
his Models/Algorithms.
9. What is difference between classic ML and DL?
FEATURE
EXTRACTOR
(f1, f2, … fk)
CLASSIFIER
SVM
Random Forest
Naive Bayes
Decision Trees
Logistic Regression
Ensemble Methods
Tom Hanks
Tom Hanks
Fixed Training
Training
~60 million parameters
10. What is difference between classic ML and DL?
CLASSIC ML DEEP LEARNING
Using optimized functions or algorithms to
extract insights from data
Using massive labeled data sets to train deep (neural) graphs that can
make inference about new data
Algorithms
- Random Forest
- SVM
- Regression
- Naive Bayes
- Hidden Markov
- K-Means Clustering
- Ensemble Methods
- More ...
Training
Data
New
Data
Inference,
Clustering, or
Classification
Untrained
model
Trained
model
CNN,
RNN,
...
Step 1: Training Step 2: Inference
Hours to Days
Milliseconds to
seconds
+ Computing
+ Labeled data
11. Sample DL application– Image Classification
Typical Training run:
•Pick a DNN design
•Input 100 million training
images spanning 1000s
categories
Test Accuracy:
•If bad: modify DNN, fix training
set or update training
parameters
13. Pain points for data scientists
FUTUREPAST
1. Use of Laptop/PC to develop their model, no storage and computing resources. Need a
server with GPU and Storage to train / validate their model.
2. Lack of knowledge to setup NVIDIA GPU Driver / Container / GPU Passthrough, etc.
3. Lack of knowledge to setup data backup, data sharing, port mapping, etc.
16. •Fuel QuAI development with GPU-accelerated computing
•QNAP NAS integrates the capabilities of a PCIe graphics
card into QTS and in-turn to Container Station
• With the power of modern graphics cards in QNAP NAS,
performance AI Modeling can be greatly boosted.
QuAI with GPU Accelerated Computing
17. QuAI Architecture
QTS 4.3.4 with GPU Driver
QNAP NAS + GPU
Container Station
Caffe MXNet TensorFlow Cuda
QuAI Containers
Graphics Card
19. Get Started with QuAI
1. Install and run
QuAI from the QTS
App Center.
2. Insert a compatible
graphics card in the
NAS.
3. Install the drivers for
the graphics card from
the QTS App Center.
4. Set GPU allocation to QTS
Mode.
5. Create the required framework
containers in Container Station
and start your first AI application.
20. Get Started with QuAI
1. Install and run
QuAI from the QTS
App Center.
2. Insert a compatible
graphics card in the
NAS.
3. Install the drivers for
the graphics card from
the QTS App Center.
4. Set GPU allocation to QTS
Mode.
5. Create the required framework
containers in Container Station
and start your first AI application.
22. Get Started with QuAI
1. Install and run
QuAI from the QTS
App Center.
2. Insert a compatible
graphics card in the
NAS.
3. Install the drivers for
the graphics card from
the QTS App Center.
4. Set GPU allocation to QTS
Mode.
5. Create the required framework
containers in Container Station
and start your first AI application.
24. Get Started with QuAI
1. Install and run
QuAI from the QTS
App Center.
2. Insert a compatible
graphics card in the
NAS.
3. Install the drivers for
the graphics card from
the QTS App Center.
4. Set GPU allocation to QTS
Mode.
5. Create the required framework
containers in Container Station
and start your first AI application.
26. Get Started with QuAI
1. Install and run
QuAI from the QTS
App Center.
2. Insert a compatible
graphics card in the
NAS.
3. Install the drivers for
the graphics card from
the QTS App Center.
4. Set GPU allocation to QTS
Mode.
5. Create the required framework
containers in Container Station
and start your first AI application.
28. Get Started with QuAI
1. Install and run
QuAI from the QTS
App Center.
2. Insert a compatible
graphics card in the
NAS.
3. Install the drivers for
the graphics card from
the QTS App Center.
4. Set GPU allocation to QTS
Mode.
5. Create the required framework
containers in Container Station
and start your first AI application.
30. •Key is building, training and optimizing your
AI Models
•Typically high performance workstations or
Public clouds are used for this process
•Generally these elements are
complemented by GPUs
Why QNAP NAS for AI?
High Performance
Workstations
Public Cloud
Platforms
But are these the Best solutions
for AI Development?
31. Why QNAP NAS for AI?
High Performance Workstations Public Cloud
•High TCO –Total cost of ownership
•Difficult to setup
•Long time to configure AI frameworks
•Not optimized to store and manage huge
data
•Very Complex pricing model
•Difficult to transfer terabytes of data to
public cloud to train AI Models
•Privacy might be an issue
32. Why QNAP NAS for AI?
Low Investments and high Gains:
Relatively less Total Cost of Ownership, compared to Workstations.
One time cost against Complex billing model of Public cloud platforms
Higher cost efficiency with QNAP NAS
Its FAST, Its EASY and its QuAI:
QuAI helps to setup AI environment in few Quick and Easy steps
Provides a state of art Wizard for quick GPU configurations.
It takes just few minutes, against few hours on workstations.
33. Why QNAP NAS for AI?
Designed to Manage Huge Data:
•Unique and most sophisticated Storage Management capabilities
with new Storage Manager.
•Provides almost Unlimited storage
•Qtier identifies hot and cold data through self-learning based on
data access rates
34. Why QNAP NAS for AI?
High IOPS:
Accelerate IOPS performance for your IOPS-intensive AI applications with
QNAP’s SSD Cache technology
35. Why QNAP NAS for AI?
Multiple layers of data protection
•Built on the security first principal to minimize the risk of data breaches with
multiple data protection mechanisms, and user privilege controls.
•Security of your Private Network thus it’s an
ideal private cloud solution for AI
•Training Data and DL Models are securely
stored within your private network.
•Best suited for research projects where
confidentiality is paramount.
36. Why QNAP NAS for AI?
QNAP NAS with most stable, robust,
comprehensive and Intelligent operating system
QTS is a strong foundation for your AI applications
and training frameworks.
37. Why QNAP NAS for AI?
Ready Availability of Huge Training Data:
Leverage Huge and diverse data which is readily available from huge range of
QNAP NAS applications to train your AI Models.
QuAI Deep Learning Model
39. NAS Models : X77
Supported Models and System Requirements
System Requirements:
QTS 4.3.4 and above
Container Station v1.8 and above
Brand Model
ASUS PH-GTX1050-2G
ASUS PH-GTX1050TI-4G
ASUS PH-GTX1060-3G
ASUS DUAL-GTX1050-O2G-GAMING
GIGABYTE GV-N1050TD5-4GD
MSI GTX1060 AERO ITX 3G OC
EVGA GTX1050 2G SC
EVGA GTX1050TI 4G SC GAMING
EVGA GTX1060 6G SC GAMING
GIGABYTE GV-N1070IXOC-8GD
NVIDIA Quadro P2000 5G
NVIDIA Quadro M2000
MSI GTX1050 TI 4GT LP
NVIDIA Quadro P4000
MSI GTX1060 6GT OCV1
Supported Graphics Cards: (TBD)
40. 1. Getting Started with QuAI
2. Image Classification using Caffe
3. Training a handwritten digits predication model using Tensorflow
4. Video Analytics using Caffe
Demo