Webinar on the Theory of Algorithms at Innopolis University

    Good day, dear readers!

    A series of our webinars last week successfully crossed the equator, and today we want to announce the upcoming webinar, and also publish transcripts of the recordings of questions from our listeners and the answers of our lecturers to them.

    So, on March 11 at 18:00 (Moscow time), our professor Manuel Mazzara will conduct a webinar on Theory of Computation. Hurry to register for the webinar at the link .



    We remind you that the collection of applications for our undergraduate program continues .

    Artificial Intelligence Webinar, Lecturer - Samir Belhauari
    Question 1:
    Professor, have you heard about the Human Brain project, whose goal is to create a model for simulating the human brain?
    Answer: Yes, the HBP project is implemented by the university in which I worked before - École polytechnique fédérale de Lausanne (EPFL-Switzerland). This project is funded by the European Union. They build a computer model of the real brain and thereby try to learn all its secrets.

    Question 2:
    What languages ​​are used today in AI programming?
    Answer: It depends on the specific industry of AI (computer vision, machine learning, search, planning, robotics, natural language processing, etc.). I would say that you should choose the language that makes it easy to create prototypes of certain elements, for example, MATLAB. However, if you plan to build a system, then C or C ++ will be the best choice.

    Question 3:
    Could you give an example of how AI can be used in the oil and gas industry?
    Answer: Artificial intelligence technologies are widely used today in the modern oil industry, including oil and gas field simulations, optimization of drilling and production, drilling automation and process control in general.

    Question 4:
    What basic knowledge is necessary for successful training in this subject?
    Answer: You will need basic programming knowledge, Probability Theory and Linear Algebra.

    Question 5:
    Does the Markov Chain relate to dynamic programming?
    Answer: Yes, it does. Markov chains are often used in solving problems of dynamic optimization.

    Question 6:
    Please tell me how much you need to understand the economy in order to develop AI mechanisms in this industry?
    Answer: You need to know exactly what the economist wants from your model in order to develop an effective model. In terms of knowledge, I believe that you only need to know basic economic terms and general concepts.

    Question 7:
    What approaches exist in the field of computer vision for applications without a priori information. What ideas exist in the field of motion detection?
    Answer: 1) The clustering algorithm will be used if there is no a priori information; 2) Motion detection works on the basis of the frame differentiation method, which compares how the intensity of pixels changes its position from frame to frame. There are two options for motion detection: a) the first method considers the change in the pattern as a whole; b) the second method considers the movement of the averaged array

    Question 8:
    What can you say about the language R? I heard that he is quite popular in the field of Machine Learning. Is there a workshop on a programming language in the course itself?
    Answer: The R language is often used, as it is free and easily accessible on the Web. I prefer to use MATLAB (we can add some tasks in the R language, if there are any). The Artificial Intelligence course will have 5 assignments, during which students will be able to use MATLAB or C ++ for coding.

    Question 9:
    Which of the following organizations - IEEE or ACM has more authority in Computer Science?
    Answer: Both of these organizations are highly reputable.

    Full webinar entry:


    Artificial Cognitive Systems Webinar, lecturer - David Vernon
    Question 1: Renat Shaikhutdinov
    How do you test complex behaviors? For example, a simple function can be tested using unit tests, but what to do in this case?
    Answer: The main problem is that cognitive systems deal with circumstances that cannot be fully detailed at the development stage, that is, it is assumed that cognitive systems are faced with changing, non-predictable and lacking integrity information. Testing these types of systems is extremely difficult. Of course, you can subject individual components to unit testing, but system testing is much more difficult. Most people solve this problem by testing the system in real conditions, observing how it interacts with living and non-living objects in enumerating possible scenarios. You can do this in the laboratory by simulating a natural environment (for example, a kitchen), or try it in a natural environment, setting it a task, and observing its behavior,

    Question 2: Victor Smirnov
    Professor Vernon, in your presentation you conceptually separate training and development. How common is this separation in cognitive science and / or robotics?
    Answer: The difference between training and development is that during training, the model provided by an external subject is adapted (calibrated), and during development the subject creates his own model. Thus, training is based on determining the parameters of the model provided by another entity, while development is based on self-creation of the model. This division is not yet so widespread in the scientific community, but it has proved to be very useful in explaining the technical methods and processes that underlie both problems. Of course, there are other ways to separate learning and development. For example, training usually focuses on one skill or knowledge, while development involves the acquisition of many skills and knowledge and understanding of their relationships. Also, learning often involves understanding how the world functions, often without taking into account the agent’s point of view on the situation. Development is always related to the attitude of the subject's abilities in the context of how the world functions. In psychology, development is the process that the subject goes through to expand his set of possible actions and extend the time interval of his ability to prospect (that is, the ability to anticipate events and the need to act).

    Question 3: Anatoly Sviridenkov
    Do you use any specific logic for knowledge? If so, what type? First order logic?
    Answer: Most cognitive systems, such as Soar and ACT-R, use a production model, that is, they are rule-based systems with conditions and related actions. Although they do not use formal logic, they are, in essence, the application of first-order predicate calculus, that is, first-order logic. Emergent (from the English emergent - emerging, unexpectedly appearing) cognitive systems do not denote knowledge by means of symbols and, therefore, do not use formal logic to justify or denote knowledge. Instead, they use commutative methods and associative techniques to present and operate on information.

    Question 4: Dmitry Chesnakov
    In your presentation, are you talking about MOOC (mass open distance learning courses) or about a course at Innopolis University?
    Answer: This is a course at Innopolis University. Professor Vernon will teach the course “Artificial Cognitive Systems” in the second semester of the 2014-2015 school year

    Question 5: Victor Smirnov
    How important is the problem of modeling phenomenal consciousness in modern cognitive systems?
    Answer: Opinions differ. Some scholars believe that there is no need to use the concept of consciousness when studying the process of cognition. Others are convinced that this is an integral part of cognition. What is exactly true is that the study of computational models of consciousness is currently a recognized area of ​​research and the study of the cognition process plays an important role. In my course, we will not address the topic of consciousness much, only when studying various types of memory (procedural and declarative memory)

    Question 6: Anatoly Sviridenkov
    What do you think of deep learning in cognitive systems? Is it possible to combine character and sub-character levels?
    Answer: Great question! Most of the scientific community believes that this is important. When I talked about cognitive architectures, I mentioned the cognitive and emergent approach. The combination of these two approaches is called hybrid cognitive architecture. There are many well-known hybrid cognitive architectures such as CLARION and ACT-R. The basic idea is precisely what you are proposing: combine symbolic and sub-symbolic forms of knowledge, that is, knowledge that is explicit and represented by a symbolic system and knowledge that is implicit and often presented using commutative techniques such as artificial neural networks.

    Question 7: Anatoly Sviridenkov
    In what area do you expect a breakthrough: brain modeling, machine learning, or artificial intelligence?
    Answer: I think in the field of machine learning, but all three areas are very important. For example, many studies are currently underway in the field of brain modeling. The results have already influenced several computing systems, such as a mirror neural system. Personally, I expect several key breakthroughs in the study of autonomous systems.

    Question 8:
    In your presentation, you pay great attention to various types of memory. Why does a machine have so many types of memory?
    Answer: We need different types of memory to encode different types of knowledge. For example, there is declarative knowledge that relates to facts about the world (metals are solid, boiling water is very hot, etc.). Another type of knowledge is skill-based knowledge, the ability to do things. This is procedural knowledge. They are encoded in a separate type of memory and require different representations when you create an artificial system. We call this procedural memory. There is also short-term and long-term memory, short-term memory stores knowledge only as long as it is necessary to carry out some task, long-term memory contains all the experience of the subject.

    Full webinar entry:


    Component-Based Software Engineering webinar, lecturer - Manuel Mazzara
    Question 1:
    Today, companies and people operate with huge amounts of data. At the same time, companies accumulate the entire history of transactions, management information and other data. People store a large amount of media data. But in real life, there are no organisms capable of storing all sensory information. Is there a paradigm in software development that addresses this issue?
    Answer: As far as I know, such a paradigm does not exist now. This area is very new now. Big Data, Data Mining, and to some extent the Deep Web are areas that are related to the issue you mentioned.

    Question 2:
    Do you think computers and machines will surpass our abilities in the future if technologies develop at such a pace? How humanity can prevent this, to always stay ahead.
    Answer: My personal opinion is no. There are certain things that only humans can do; machines can never make them. The speed of technological development and the speed of machines are not related to this. There are separate thought processes that machines are simply not able to have. For example, computability and the Touring-Church hypothesis. Naturally, this is just my opinion. However, there is no evidence to the contrary. Some of my colleagues are sure that in 50 years we will have machines with consciousness. I'm not sure if this will happen.

    Question 3:
    As far as I understand, we use component-based software engineering daily. Is there any difference between CBSE and traditional programming techniques?
    Answer: The need for decomposition of software into smaller components exists not only in the last 10 or 15 years, but much longer. At the very beginning of the development of computer science, it became obvious that big problems are more effectively solved if they are divided into small parts. CBSE is certainly part of Software Engineering. CBSE is an evolutionary development of previous concepts, such as structured programming and object-oriented development.

    Question 4:
    Which IT development sectors are most important these days and will be most important in the near future?
    Today, cloud computing, Big Data, and Data Mining are much discussed. With the advent of social networks, we have noticed tremendous commercial interest in these areas (from Facebook, Twitter, etc.)
    Answer: Another important aspect is the field of electronic medicine, the technologies in which are used to improve the quality of life of people, especially the elderly and people with disabilities. We can help these people thanks to modern technology. For example, this is implemented through mobile applications and devices. Personally, I plan to pay special attention to these areas in the future.

    Question 5:
    What do you think of truly decentralized systems? For example, systems in which all links are interchangeable?
    Answer: This question is very interesting. Such systems exist, but we will not discuss them during our course. I am ready to discuss this issue extracurricularly.

    Question 6:
    Do you think that knowledge is not as important as the ability to learn and analyze?
    Answer: I personally believe that the most important is the ability to learn "how to learn." Therefore, for me, knowledge is less important than the ability to use mental constructions in work. You can always get knowledge on the Internet, in databases, in other places; but you must be able to process information. If you do not know how to work with information, then you will never understand many things. For me, teaching methods are more important than just knowledge.

    Question 7:
    Is there a difference between small and large software engineering projects? Do we need to apply different approaches to them?
    Answer: There are two answers to these questions. Of course, there is a difference. The need for object-oriented development arose due to the complexity of managing large projects. For small projects, we do not see such a need. The second answer: during classes, we will not be able to work with large projects, and we will apply the above methods to small projects. Of course, small projects are not always as well suited to this goal. However, the final answer is that we can apply these techniques for both cases.

    Question 8:
    What do you think is the most important event in the field of computer science over the past 10 years?
    Answer: From a scientific point of view, there have been many discoveries, for example, in the field of algorithms. However, they are not so important to the general public. The most visible and essential part, of course, I consider the spread of social networks. They and mobile technology are the most visible changes of the past decade. Twenty years ago, no one could have imagined that people would sit in a restaurant and pay more attention to their mobile device than to eating and communicating with living people.

    Full webinar entry:


    Machine Learning Webinar, Lecturer - Samir Belhauari
    Question 1:
    Over the past few years, many software libraries and environments have been created for researchers in the field of Machine Learning. However, in the field of hardware, we see only a few examples of special processors for digital data processing and machine learning. What are the most significant developments you could mention in recent years?
    Answer: I worked on a project on the use of sensors and chips for gas recognition and detection based on Gaussian processes and a model of a mixture of normal distributions for machine learning. SVM algorithm was applied in the form of hardware software .

    Question 2:
    What do you think of skin cancer? Mortality rates from this species are lower than from other dangerous types of cancer, while it is most suitable for machine recognition due to its visual symptoms.
    Answer: Breast cancer and lung cancer are most common among women and men, and regarding skin cancer there are many studies regarding segmentation and diagnosis.

    Question 3:
    You mentioned a few subjects that you need to study before starting Machine Learning. Could you advise subjects that can be studied in parallel or after Machine Learning classes?
    Answer: I will “refresh” students' knowledge of mathematics and algorithms before the start of our course, and I will also provide additional materials that will help to better understand the basic concepts of the course.

    Question 4:
    What projects will students do during the training?
    Answer: Students are required to know the basics and algorithms of ML, as well as rewrite (understand) codes

    Question 5:
    As I understand it, the main goal of the course is to let the machine learn without programming through the generalization process.
    Answer: it may or may not be programmed, as some algorithms are dynamic, but some are not.

    Question 6:
    Do you think that generalizing the information obtained is enough to create artificial intelligence?
    Answer: I believe that this is not enough, as some applications require the use of innovative algorithms without data

    Question 7:
    Some universities note that some universities have succeeded in programming robots and computers to summarize data. Is it possible to teach machines to make decisions, aware of their consequences in the future?
    Answer: Of course, it is possible to teach computers to make the best decisions and conclusions, however, to accomplish this task it is important to model our problem in mathematical terms.

    Question 8:
    What literature would you recommend for beginners on this topic? What areas of mathematics are important for learning machine learning?
    Answer: You need to know the following areas:
    • Basic probability, matrices and calculus
    • Familiarity with some programming language C / C ++ and MATLAB
    References:
    S. Russell and P. Norvig. Artificial Intelligence: A Modern Approach, 3rd Edition, Prentice Hall (2010).
    D. Koller and N. Friedman. Probabilistic Graphical Models, MIT Press (2009)
    R. Sutton and A. Barto. Reinforcement Learning: An Introduction, MIT Press (1998)
    E.Tsang. Foundations of constraint satisfaction. Academic Press (1993)

    Full webinar entry:

    Also popular now: