Tim:

Branislav Vezilić E2 53/2016
Nemanja Rašajski E2 2/2016
Nina Marjanović E2 43/2016

Opis:

Planiranje putanje kretanja autonomnog, agilnog vozila u dinamičnom, simuliranom okruženju. Projekat treba da reši problem generisanja i izvršavanja plana kretanja autonomnog vozila. Za polu-nadgledano obučavanje konvolutivne neuronske mreže biće korišćeni podaci dobavljeni iz simulacije, a pronalaženje dobrog i optimalnog rešenja zavisiće od genetskog algoritma.

U zavisnosti od vremena, projekat može progresivno postajati kompleksniji uvođenjem prepreka, saobraćajnih znakova, semafora.

Description

Generate company logo for the given parameters (business area, name of the company).
Planed usage of Generative Adversarial Nets.
Haven't found appropriate dataset yet - in the worst case it will be gathered manually.

Tech

Language: Python

Team

Zoran Luledzija E2 93/2016
(Working alone)

Smart Alerting

Goal of this project is to make neural network model that will detect type of anomalies on Apache Cassandra nodes. Slow queries on Apache Cassandra nodes are caused by some anomalies in CPU, Memory, Disk, Network etc. This project will compare results of convolutional 1D and recurrent neural network.

Dataset

I will use labeled data. Labels represent type of anomaly. Monitoring machine will collect time series metrics on Cassandra nodes and store data in database. This data will represent normal type of data. Then, I will use tools to cause anomalies on cluster (cause high cpu usage on nodes and that will get metric with label cpu anomaly. Same is for memory, disk etc).
After collecting dataset, next task is to prepare data for training. Some preprocessing steps will be feature extraction, mean normalization etc.

Optional Task

• If I have time untill 16.01.2017
  Train model with unlabeled data that will detect just if anomaly happened or not. Value of this task will be constant learning on streaming data, so model will always be „fresh“ and will follow behavior of cluster. And if anomaly is happen, model will send email alert with timestamp and current values of metrics and cause of anomaly (original task). On this task I plan to use autoencoders and/or GAN.

Tech

Project will use: Scala (optional Python), TensorFlow, Apache Spark, Apache Cassandra

Working alone on this project: Stanko Kuveljić E2 20/2016

Opis projekta:

Implementirati klasifikaciju polova nad IMDB dataset-om i uporediti rezultate nad pretreniranim VGG16 i ResNet50 modelima.

Realizacija projekta u programskom jeziku Python.

Samostalni projekat: Nevena Simić E2 94/2016

Description:
The goal of the project is the enabling/improvement of navigational capability of an agent. In a 3D environment made up of multiple interconnected rooms and corridors the agent has to find its way to its given objective. In the realization, the OpenAI Gym toolkit will be used, also the DOOM environments from OpenAI Gym, specifically the DoomMyWayHome-v0. The theme of the project is influenced/based on the task/assignment of the same name on OpenAI Gym.

Tech:
Language: Python

Team:
Gabor Vadocki E2 43/2016
Working alone on the project.

Description

The Digital Mammography DREAM Challenge will attempt to improve the predictive accuracy of digital mammography for the early detection of breast cancer. The primary benefit of this Challenge will be to establish new quantitative tools - machine learning, deep learning or other - that can help decrease the recall rate of screening mammography, with a potential impact on shifting the balance of routine breast cancer screening towards more benefit and less harm. Participating teams will be asked to submit predictive models based on over 640,000 de-identified digital mammography images from over 86000 patients, with corresponding clinical variables.

Tech

Project will be deployed as a Docker image, as it is required by the challenge.
Language: Python
Libs: TensorFlow or Keras
Model: Convolutional Neural Networks

Team

Aleksandar Lukić (E2 97/2016)
Opened for new members.

Članovi tima:
- Milan Keča E2 67/2016
- Dragan Vidaković E2 66/2016

"Ako vas nema troje, da li prihvatate da vam se neko pridruži?" - Možda :)

Opis projekta:
Na osnovu definisanog zadatka (na engleskom jeziku) neuronska mreža će generisati Python source code koji će rješavati definisani zadatak. Podaci su Python source code-ovi, skupljaju se manuelno, i biće dostupni ostalim, zainteresovanim timovima. Planirana je upotreba rekurentnih neuronskih mreža i GAN-ova (ako to bude potrebno).

Валидација људског потписа на основу слике коришћењем конволутивних неуронских мрежа. Сет података за обучавање је : ICDAR 2011 SigComp. Чланови тима: Петар Стошић E244/2016, Небојша Поповић E245/2016.