Deep Learning Fundamentals

Brief description

The training program is intended for adult employees who do not have an
appropriate education in the field of computer or electrical engineering or do have such education but have limited or no prior exposure to data science, machine learning and deep learning. The training program is designed for those in need to understand, utilize, and apply (re-train or fine-tune) deep learning models in their daily work. The program covers theoretical and practical introduction to modern deep learning and its primary areas of applications. To keep the program focused, the lectures are primarily oriented towards applications in computer vision, since those applications are among the most common and ubiquitous in modern industrial applications. The focus is on developing skills required to understand:
i) problems addressed by modern machine learning, and
ii) methods and tools used to solve those problems.
The program will provide participants with knowledge and skills required to use (but also design) various types of neural networks, such as: shallow feed-forward networks for basic regression and classification tasks, convolutional networks for image processing (classification and object detection), as well as anomaly detection. The program is envisioned in such a way that the acquired knowledge can be directly applied to solve a variety of practical problems, but that it can also be easily extended and upgraded through other courses and individual work.


Novi Sad

60 hours Total
30 Theoretical
20 Practical

1200 Euros


University of Novi Sad, Serbia, 2007-2011
PhD in Electrical and Engineering and Computing
Optimal and Sub-Optimal Control of a Class of Heat Diffusion Processes

BSc + MSc
University of Novi Sad, Serbia, 2001-2006.
Integrated studies in Electrical Engineering and Computing

Professional Experience:

University of Novi Sad
Faculty of Technical Sciences, Department of Computing and Control.
Full Professor, 2021 to present.
Associate Professor, 2016-2021
Assistant Professor, 2011-2016
Teaching assistant 2006-2011.

Visiting Professor and Scholarships

Visiting Professor at Bordeaux IMP, 2023
Teaching and mentoring students on Master level in the field of Control and applications of ML in Control (Reinforcement Learning in particular).

Objectives of the program

The goals of the program are to train people to:

Who Should Participate?

The target group of training are employed adults who have completed secondary vocational school or secondary professional education, or higher education in an engineering field, but had no or limited previous experience with data science, machine learning, or deep learning.

Course in details

– Short history of deep learning and artificial intelligence (AI), machine learning (ML), and deep learning (DL).
– Overview of the main applications of DL
– Introduction to neural networks: nodes, activation functions, connections, weights, biases, layers

– Python language fundamentals.
– Working with numerical computing library “numpy”.
– Working with plotting library “pyplot”.
– Working in interactive programming environment “Jupyter”.
– The concept of virtual environment, and installation of packages in isolation.

– Overview of fundamental problems of machine learning: supervised, unsupervised, and reinforcement learning.
– Data collection and pruning of data. The concept of outliers.
– Model selection, training, validation, and testing.
– Data partitioning
– Linear classification and regression
– Nonlinear classification and regression

– PyTorch tensors and basic operations
– Computation graphs
– Automatic gradient and backpropagation
– GPU support
– EXAMPLE: Solving a simple classification problem using PyTorch

– Built-in activation functions and their properties
– Built-in optimizers and their properties
– Initialization of DL models
– EXAMPLE: Solving a more complex classification problem using PyTorch

– Representation of images. Images as tensors
– Convolution and its properties
– Convolutional layers, convolutional neural networks (CNNs)

– GoogleNet/Inception architecture
– ResNet architecture
– Transfer learning
– EXAMPLE: Solving a benchmark image classification problem (e.g. CIFAR10)

– Latent space representation
– Encoder-decoder networks
– Convolutional encoder-decoder networks
– Application to anomaly detection

– Object detection and image segmentation – problem description
– Dataset preparation for object detection tasks. COCO and YOLO annotation formats
– Faster R-CNN
– YOLO networks (the original YOLO architecture and its most important modifications, including YOLO v3 and YOLO v5)
– EXAMPLE: Solving an object detection problem

Team assignment in which all participants will be split into 3-5 person teams, and given practical computer vision problems (each team will be given a separate problem, e.g. a specific image classification, object or anomaly detection tasks) and will be required to produce working solution in the given timeframe.