BDV Webinar Series - Lara - Deep Learning for Everybody

DEEP-Hybrid-DataCloud is funded by the Horizon 2020 Framework Programme
of the European union under grant agreement number 777435
DEEP as a Service:
Deep Learning for everybody
Lara Lloret Iglesias
<lloret@ifca.unican.es>
IFCA (CSIC-UC)

DEEP-Hybrid-DataCloud
 Designing and Enabling E-Infrastructures for intensive data Processing in
a Hybrid DataCloud
 H2020 project, EINFRA-21 call
 Started November 1
st
2017
 Global objective: Promote the use of intensive computing services by
different research communities and areas, and the support by the
corresponding e-Infrastructure providers and open source projects
https://deep-hybrid-datacloud.eu/

DEEP Partners
 9 academic partners
 CSIC, LIP, INFN, PSNC, KIT, UPV, CESNET, IISAS, HMGU
 1 industrial partner
 Atos
 6 countries
 Spain, Italy, Poland, Germany, Czech Republic, Slovakia

 Provide specialized cloud services to develop, exploit and share
machine learning applications (machine learning, deep learning)
 Covering the entire application development cycle
 Transparent access to specialized computing resources
(accelerators, supercomputing systems) → lower entry barrier
 Build a catalog or marketplace as an application exchange
method → ease of use, collaboration, dissemination of
knowledge
Project focus

 The development of a model starts on a PC
 The model is trained with training data (labeled or not) and tested with a different data set
 This phase requires an adequate computing infrastructure
 The model is evaluated with data that you have never seen before.
 The cycle is repeated until the evaluation is satisfactory
 The work is published (academy): architecture, data configuration, scientific article, etc.
 Or it is deployed in production (industry)
Open issues:
 Access to infrastructure?
 How to share and exchange knowledge?
 Is there a standard way to deploy it as a production service?
 Good practices?
 And many more ...
Development cycle in Machine Learning

Training a Machine Learning/Deep Learning model is a very complex and
computationally intensive task requiring the user to have a full setup involving a
certain hardware, the adequate drivers, dedicated software and enough
memory and storage resources.
Very often the Deep Learning practitioner is not a computing expert, and want all of
this technology as accessible and transparent as possible to be able to just focus
on creating a new model or applying a prebuild one to some data.
The DEEP-HybridDataCloud project offers a framework for all users, and not just for a
few experts, enabling the transparent training, sharing and serving of Deep Learning
models both locally or on hybrid cloud system.
The goal: Deep as a Service

The DEEP-HybridDataCloud solution is based on Docker containers packaging already
all the tools needed to deploy and run the Deep Learning models in the most
transparent way.
One doesn’t need to worry about compatibility problems, everything has already been
tested and encapsulated so that the user has a fully working model in just a few minutes.
Using Docker enables you to package your entire application with all of its dependencies with
zero headache for compatibility issues or machine dependency.
Docker containers

The DEEP Open Catalog provides the universal point of entry to all services offered by DEEP
https://marketplace.deep-hybrid-datacloud.eu
Its offers several options for users of all levels to get acquainted with DEEP
DEEP Open Catalog
 Basic Users can browse the DEEP Open Catalog, download a certain model and
apply it to some local or remote data for inference/prediction.
 Intermediate Users can also browse the DEEP Open Catalog, download a model
and do some training using their own data easily changing the parameters of the
training.
 Advanced Users can do all of the above. In addition, they will work on more
complex tasks, that include larger amounts of data.

API accessible from web browser
To make things even easier, we have developed an API allowing the user to interact with
the model directly from the web browser.
It is possible to perform prediction, train or check the model metadata just with a
simple click!
Let see how this work!

/
Success stories: Phytoplankton

 This use case is concerned with the classification of phytoplankton images
 Tensorflow example implementation
 The network used is a Xception
Phytoplankton Use Case

Browse the marketplace

Prediction: Image classification

Prediction: classification probabilities
The output is given in JSON format that can be very easily integrated with any other
application needing to access the results.

Training: image classification

DEEP OC Image Classification (Tensorflow)
Training: image classification

Mechanism to establish collaborations between private companies and public
sector to access technological services, research data and human capital
https://www.eosc-hub.eu/digital-innovation-hub/
 DEEP and EOSC-DIH have established a mutual collaboration
 DEEP provides services to DIH
 DIH seeks SMEs to conduct application pilots
 DIH, through the EOSC-Hub project, will also provide resource providers
DEEP is included as a provider of the following services in DIH:
ML application porting to EOSC technological infrastructure
ML implementation best practices
AI-enabled services prototyping in EOSC landscape
EOSC Digital Innovation Hub (DIH)

ML application porting to EOSC technological infrastructure
ML implementation best practices
AI-enabled services prototyping in EOSC landscape
 Access to high performance computing services (GPUs, HPC) for training in AI
applications, machine learning and deep learning → access to infrastructure
 Access to services for automatic deployment of models as a service
(DEEPaaS) → global and scalable access to the deployed model
 Publication of applications in the project catalog → impact and visibility
EOSC Digital Innovation Hub (DIH)
It is possible to grant sponsored access to resources

 The DEEP-HybridDataCloud approach offers a user friendly entry point and a set of open
tools to take advantage of the latest DL/ML machinery in an easy transparent way
 Machine learning service provider at EOSC
 We cover all the steps of the machine learning application development cycle
 Tools for federated learning: ensuring integrity, privacy, reliability, insurance, auditability,
traceability...
 Great possibilities for exploitation by the industry through the EOSC DIH (pilot
development)
 Possibility of access to computing and storage resources (sponsored through the pilot)
 Possibility of technical support, both by the EOSC and by DEEP Possibility of using a
combination of services of the DIH portal and the EOSC
 Increase in visibility, publicity and dissemination of results
Conclusions

References
Contact:
E-mail: aloga@ifca.unican.es, lloret@ifca.unican.es, deep-po@listas.csic.es
Web page:
https://deep-hybrid-datacloud.eu
Social media:
https://twitter.com/DEEP_eu
DEEP Open Catalog:
DEEP documentation:
http://docs.deep-hybrid-datacloud.eu
DEEP YouTube channel:
https://www.youtube.com/playlist?list=PLJ9x9Zk1O-J_UZfNO2uWp2pFMmbwLvzXa

DEEP-Hybrid-DataCloud is funded by the Horizon 2020 Framework Programme of the European union under grant agreement number 777435
Thank you
Any Questions?Thank you!

DEEP-Hybrid-DataCloud is funded by the Horizon 2020 Framework Programme of the European union under grant agreement number 777435
Thank you
Any Questions?
Services at the EOSC portal

BDV Webinar Series - Lara - Deep Learning for Everybody

More Related Content

Similar to BDV Webinar Series - Lara - Deep Learning for Everybody (20)

More from Big Data Value Association (20)

Recently uploaded (20)

BDV Webinar Series - Lara - Deep Learning for Everybody