The “Artificial Intelligence” Master’s program encompasses a wide range of courses that cover both foundational and advanced topics.
The "Deep Learning" course aims to provide students with comprehensive knowledge of complex machine learning methods, focusing on deep neural networks, their design, tuning, applicability and use scenarios. The lecture approaches the topic of deep machine learning models in an applied manner, considering the various aspects involved when implementing such models in popular computational frameworks. Topics include various types of deep neural network layers and architectures: convolutional and recurrent networks, attention-based models, models adapted to time-series data, and transformers. Strongly-related topics such as dataset management and computational models for deep neural networks are also considered.
Course description
The "Fundamentals of Machine Learning" course aims to provide students with a comprehensive understanding of the core concepts, techniques, and applications within the field of machine learning. The lecture aims to offer a structured exploration of key topics such as supervised and unsupervised learning, model training and evaluation, feature engineering, and the underlying algorithms. Students are expected to gain significant knowledge and insight into the field of machine learning, enabling them to apply the corresponding algorithms and techniques to real-world problems.
The course introduces the fundamental concepts of Knowledge, Representation, and Reasoning. It continues with the language of first-order logic, knowledge expression, and resolution: syntax, semantics, facts, handling variables, and quantifiers. Knowledge-Representation surveys the spatial and featural alternatives, network models, structured representations, mental models, general and specific information in representation, and their practical use. Ontologies refer to a manner of organizing related concepts. This section reviews the main components, types, languages, tools, and methods for organizing ontologies and knowledge extraction. Automatic theorem proving focuses on resolution, model checking, and first-order theorem proving. Non-monotonic reasoning discusses default reasoning, autoepistemic logic, and circumscription.
The goal of this course is to present the general issue of Natural Language Processing while learning theoretical concepts and acquiring practical skills. Over the course of the semester, current tools and technologies in the field of NLP will be reviewed, while stimulating research skills in this field. One objective is to investigate the applications of NLP across various domains, such as text classification, similarity analysis, text generation, and machine translation. Students are expected to comprehend and familiarize themselves with the theoretical principles while also developing practical skills in the field. Key concepts that will be addressed include: text preprocessing and text representation; word embeddings; attention mechanisms and transformer networks; text classification; text similarity; text generation; question answering and machine translation, etc.
The “Probabilistic Reasoning” course main purpose is to obtain the necessary knowledge to understand and solve different problems using probabilistic reasoning methods and the ability to work with these concepts. The lecture aims to provide a structured exploration of key topics in the field such as design and development of Bayesian network, junction tree algorithm, inference in dynamic Bayesian networks, Gaussian mixture or expectation maximization algorithm. The students will gain the knowledge to implement the acquired concepts for elaboration of interesting applications.
Course description
Course description
The “Big Data Techniques” course is designed to provide students with the skills and knowledge needed to develop applications and techniques for the processing of large-scale data sets. The course covers a wide array of topics, including MapReduce techniques, data acquisition and retrieval, algorithm design for big data processing, pattern identification, statistical data processing and data analytics. Additional related areas of interest covered by the course are data compression, clustering and classification methods, stream processing, and big data visualization.
The primary goal of this course is to provide the students a clear view of how the brain proceses information and learns. The secondary aim is to present the existing neuromorphic chips and platforms as well as their advantages. The students will be able to model the behavior of independent neurons and of the neural structures for particular applications starting from image and sound processing to motion control. In addition important aspects related to modelling of superior processes of the brain and advanced mechamisms for learning at the synapse level will also be presented. All implementations of the artificial neural strutures will be done using software simulations.
This course provides in-depth knowledge of Cloud Computing, covering key topics such as Cloud architecture, service models, and architectural aspects. Participants will learn about Cloud storage systems and gain practical skills in Cloud application development. Topics also include Cloud infrastructure, resource management, and serverless applications. Additionally, the course covers notification services, virtualization, and open-source solutions like OpenStack and OpenShift with Docker and Kubernetes. Students will explore storage systems, databases, and Cloud security, providing a comprehensive understanding of the Cloud computing landscape.
The "Computer Vision" course provides a rigorous exploration of computer vision, integrating both fundamental principles and cutting-edge techniques. Topics include image and video restoration, color models, texture analysis, and image enhancement. The curriculum covers feature extraction, segmentation methods, and traditional object detection. Advanced sections focus on deep learning applications in computer vision, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), generative models, and attention mechanisms. The course emphasizes practical applications, supported by relevant datasets and benchmarks, equipping students with the skills to tackle contemporary challenges in the field.
The "Data Analytics" course prepares students with the fundamental skills and knowledge necessary for effective data interpretation. Beginning with an overview of the field, the course explores a range of techniques for inspecting, cleaning, transforming, and modeling data. The primary objective is to uncover valuable insights, draw informed conclusions, and support decision-making processes. The curriculum covers various methodologies, including statistical parametric and nonparametric tests, which provide a quantifiable degree of confidence in data-driven decisions. It also addresses anomaly detection to identify unusual patterns that deviate from expected behavior. Special emphasis is placed on time series analysis, demonstrating techniques to extract meaningful statistics and identify patterns, trends, and seasonal variations.
Course description
Course description
This course aims to address the general issues encountered when solving optimization problems. It will review specific terminology, concepts, and mechanisms that can be applied to various problems, considering the practical limitations encountered in real-life situations. Stochastic and deterministic approaches and bioinspired metaheuristics (such as genetic algorithms, differential evolution, ant colony optimization, etc) will be discussed in the context of unconstrained and constrained applications. Strategies to combine mechanisms for diversity control and performance enhancement, such as hybridization, parameter adaptation, global-local search combinations, etc., will be analyzed and discussed from multiple perspectives (complexity, resources consumed, efficiency).
The “Explainable and Neuro-Symbolic Learning Methods” course offers a broad exploration of knowledge and practical applications specific to explainable AI and neuro-symbolic systems. This course delves into the fundamental concepts of XAI, focusing on model-based, input-based, and output-based explanations using techniques like saliency maps, rule extraction, and model visualization. Students will gain insights into model-agnostic algorithms such as LIME and SHAP, and explore example-based explanations including counterfactuals and adversarial examples. The course also covers hybrid neuro-symbolic systems and their applications, and introduces cognitive architectures such as ACT-R, SOAR, and SPA.
The “Multiagent Systems” course offers a broad exploration of knowledge and practical applications specific to multiagent systems. This course delves into the fundamental concepts of multiagent systems, focusing on how agents interact, cooperate, and coordinate to solve problems in distributed environments. Students will gain insights into various agent architectures, including logic-based, reactive, belief-desire-intentions (BDI), and layered architectures. Topics cover modelling agent rationality with game theory, inter-agent communication, search algorithms tailored for multiagent systems, and mechanisms such as auctions, voting, bargaining, and contract net protocols. The course also addresses coordination methods, learning within multiagent systems, and introduces students to multiagent development frameworks.
Course description
The “Reinforcement Learning” course offers a broad exploration of knowledge and practical applications specific to reinforcement learning. This course delves into the fundamental concepts of reinforcement learning, focusing on how agents learn optimal behaviors through interactions with their environment. Students will gain insights into various reinforcement learning algorithms, including Markov decision processes, temporal-difference learning, and policy gradient methods. Topics cover planning and learning with tabular methods, on-policy and off-policy prediction with approximation, and model-based reinforcement learning. The course also addresses advanced techniques like n-step bootstrapping algorithms, eligibility traces, and introduces students to state-of-the-art reinforcement learning frameworks.
Course description
Course description
Course description
"This semester-long project provides students with hands-on experience in applying Edge AI to solve a real-world problem. The goal of this project is to encompass various aspects of the Artificial Intelligence Master’s curriculum, including algorithm selection, data processing, model development, and deployment on edge devices. Key concepts: edge AI, technology stack for edge AI (ONNX Runtime, TensorFlow Lite, TFLite Micro, ML Kit for Firebase, PyTorch Mobile, Core ML, Embedded Learning Library -ELL, Apache MXNet, etc), SoC Platform, Nvidia Jetson, Raspberry Pi, Edge TPU."
While technological evolution has been beneficial to more robust and secure applications, it has also allowed a greater variety of cyber-attacks. Classic security tools tend to become obsolete, while a huge amount of both benign and malicious data is seemingly unused. Nowadays, AI/ML techniques seem to be the de facto standard in processing such data and cyber-security seems to be yet another field they could be beneficial for. The Intelligent Cyber-Security course focuses on both the benefits and drawbacks of using AI/ML in cybersecurity through three practical case studies: host, network and social attack patterns. Adversarial ML threats are also presented, alongside the threat posed by the malicious misuse of AI/ML tools.
This project-based discipline focuses on developing practical skills in visual intelligence through team-based projects. Students will work in pairs to propose, design, implement, evaluate, and present solutions to real-world problems. The discipline covers project proposal, literature review, prototype development, and final presentations. Students will utilize advanced hardware and software resources such as HTC Vive, NAO robot, OpenCV, PyTorch, and TensorFlow. The hands-on approach aims to foster collaboration, critical thinking, and technical expertise, preparing students to tackle contemporary challenges in the field of visual intelligence.
The "Voice and Speech Recognition" course presents an overview of different techniques and methods that interact with the field of voice analysis. It has evolved from the signal processing field, by analyzing waveforms and spectrograms, and is now actively developing even with neural networks. The course goes over Phonetic concepts, illustrates ASR systems (Automatic Speech Recognition), and delves into acoustic modelling, with algorithms such as HMM (Hidden Markov Models) and ideas based on MFCC (Mel-frequency cepstral coefficients). Multiple recent ASR approaches from the industry are presented, along with foundation models and TTS concepts (Text-to-Speech).
Legal and Religious Holidays:
30.11.2023, 01.12.2023, 25.12.2023 , 26.12.2023, 01.01.2024. 02.01.2024, 24.01.2024 01.05.2024, 03.05.2024, 06.05.2024, 01.06.2024, 24.06.2024, 15.08 2024.
Practical Activities (3 -6 Weeks)
will be scheduled either during the entire academic Year
Gheorghe Asachi Technical University of Iasi, with the oldest engineering education tradition in Romania, is dedicated to developing technology for the community through research, teaching, and technology transfer.
Copyright © 2024 TUIASI. All rights reserved.
