Propostas para Dissertação

Mestrados no Departamento de Informática



Consultar ficha completa de uma proposta de dissertação

proponente: António Esteves
instituição/empresa: Universidade do Minho - Departamento de Informática
tema/título: Apply Caffe2 framework to deliver AI-powered mobile applications
área científica: Engenharia de Software
local: DI-UM
curso de mestrado: Mestrado Integrado em Engenharia Informática
descrição:
The objective of this proposal is to develop mobile a application embedded with intelligence powered
by deep learning models [1][2]. For this purpose it is suggested the use of Caffe2 framework
[3][4]. Intelligent features can be expressed for example on smart image manipulation, smart
interaction, recommendations, deep personalization, opinion, and intelligent answers. The
intelligence provided by deep learning models, which learn essentially with historic data, can and
should be enhanced by context information. Context information is essentially provided by sensors,
such as GPS, photographic camera, microphone or touch screen. In April 2017 Facebook made Caffe2 open source. Caffe2 is a simple, flexible and cross platform deep
learning (DL) framework. Built on the original Caffe, Caffe2 is designed with expression, speed,
and modularity in mind. Caffe2 comes with native Python and C++ APIs that work interchangeably.
Thus we can prototype quickly in Python and optimize in C++ later. Caffe2 is tuned to take
advantage of the latest NVIDIA Deep Learning libraries (cuDNN, cuBLAS, and NCCL), to deliver
high-performance, multi-GPU acceleration for desktop, data centers, and mobile devices. The
framework adds deep learning \"intelligence\" to mobile and low-power devices by enabling the
programming of iPhones and Android devices. Complex DL models can be trained on desktops and
deployed on mobile devices. There is also available a Caffe2 collection/zoo of pre-trained models that can be run with just a
few lines of code. Qualcomm, the company that produces the processors that power a significant part of mobile phones,
is also collaborating with Facebook to optimize Caffe2 [9]. Qualcomm has its own neural processing
engine (NPE) framework. The NPE is designed to do the heavy lifting needed to run neural networks
efficiently on Snapdragon. For guiding the work to be developed, it is suggested looking at the AICamera application [5], which
was built with Android Studio and Caffe2 library. Guidelines to integrate Caffe2 on iOS or Android
are available on [6]. Tools such as Facebook WIT.AI and Google API.AI are also interesting resources in helping develop
intelligent applications. References: [1] Accelerating Convolutional Neural Networks for Mobile Applications. Peisong Wang and Jian Cheng,
2016. [2] MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. A. Howard,
M. Zhu, et al., 2017. [3] https://caffe2.ai [4] https://github.com/caffe2/caffe2 [5] https://caffe2.ai/docs/AI-Camera-demo-android [6] https://caffe2.ai/docs/mobile-integration.html [7] Caffe2: The Deep Learning Framework for Mobile Computing, April 2107. https://fossbytes.com/caffe2-the-deep-learning-framework-for-mobile-computing/ [8] Delivering real-time AI in the palm of your hand. Yangqing Jia and Peter Vajda, Nov 2016.
https://code.facebook.com/posts/196146247499076/delivering-real-time-ai-in-the-palm-of-your-hand/ [9] Caffe2 and Snapdragon usher in the next chapter of mobile machine learning. Qualcomm, April
2017.












qualcomm.com/news/onq/2017/04/18/caffe2-and-snapdragon-usher-next-chapter-mobile-machine-learning


Voltar...