Department of Intellectual Systems of the RSUH: the teacher’s view
My previous post on the use of the free statistical package R in teaching statistics caused a very useful (at least for me) discussion. However, one of the participants suggested that I describe my teaching experience in a wider context. This post is an attempt to summarize your 15 years of teaching experience in the department of intelligent systems of the RSUH.
Currently, in several universities in Moscow there are departments and departments that teach the specialties “Artificial Intelligence”, “Intellectual Systems”, etc. Among them are the Department “Mathematical Theory of Intellectual Systems” of the Faculty of Mathematics of Moscow State University, Department No. 29 “Control Intellectual Systems” and № 22 "Cybernetics" MEPI, the basic department "Intellectual systems" FUPMa MIPT in the EC of the Russian Academy of Sciences ...
Less well-known is the separation of intellectual systems (in the humanitarian sphere) of the RSUH. The ideological inspirer of the creation of this department and its leader is Dr. Konstantinovich Finn , Professor. By the will of fate, I was lucky to learn from him and work under his leadership at the VINITI RAS and at the department described. The principles underlying the department were freely discussed by V.K. Finn and his staff, so I hope my statement will be close to the original. I will make a reservation that this is my statement, so it will be as I understand it.
')
Now, in connection with the transition of education to the system of "bachelor-master" curriculum will change, so in my presentation I will try to reflect the improvements (and deterioration) introduced in the educational process by this reorganization. In 2008, the OIS of the RSUH won the tender of the Federal Agency for Education of the Russian Federation to create training programs in the specialty “Intellectual systems in the humanitarian sphere”. Therefore, the described principles are supposed to be implemented in all universities of Russia, teaching students in this specialty.
Ideologically, the teaching of IHVS specialty is divided into several blocks of academic disciplines that go through all the training.
The first is applied mathematics:
The second block is logic:
The third is programming:
Fourth - intellectual technologies:
Of course, there are a number of mandatory disciplines in the humanities cycle.
A large number of disciplines are related to the variable part (they can be chosen by students): genetic algorithms and neural networks, fuzzy sets, theory of random processes, the theory of similarity, parallel programming, ontologies and knowledge representation, etc. Many of these courses at RSUH are taught by involved experts.
Historically there is a large block of linguistic disciplines: morphology, syntax, semantics, lexicography, computational linguistics. Due to the reduction of the load on the student, in the new plans this part will be significantly reduced.
However, the ideality of such a curriculum construction faces real difficulties. I consider the main one to be a lack of understanding by applicants of where they go, what subjects are basic, what they will become after graduation from the university. Another difficulty is the inconsistency of training courses.
I will try to explain it on a personal example. I taught and teach the courses "Axiomatic Systems and Elements of Model Theory", "Machine Learning", "Logic Means of Intelligent Systems".
Since the last course depends on the first one, as well as on the basic course “Mathematical Logic”, which is read by V.K. Finn in the first two semesters, I managed to harmonize the material so as to avoid duplication (by organizing, of course, a reminder of the material). I must say that students having studied the method of analytical tables on the 1st course, most of them can use it on the 5th year when they study “Logical tools ...”.
With the course "Machine learning" the situation is radically different. Now my course (under the influence of my personal scientific interests) is shifting towards the probabilistic theory of learning (ala VN Vapnik). But the course of the theory of probability and statistics does not imply the presentation of the material I need (Chernov's inequality, the method of re-sampling, etc.). Therefore, I have to interrupt the presentation of the material by inserts from probability theory. And still, as I believe, students do not understand. Too much different probabilistic material from the algorithmic. I also observe the fact that students forget about NP-completeness (this is needed to prove that sometimes the learning algorithm does not exist). This is aggravated by the fact that the courses “Theory of Algorithms” and “Theory of Relational DBs” interpret it in different terms. Finally, about the theory of information on Shannon (needed for the presentation of the ID3 algorithm of learning decision trees), students do not know anything at all!
This year I even decided to give a course in mathematical statistics, coordinating it with the course of probability theory. I am afraid, however, that this will help me only in a year, when third-year students will start to learn Machine Learning.
As for the wrong goal-setting and inability to single out the most important things for applicants and students, I can only add that I observe a diligent study of mathematical analysis, which they safely forget about course 5, because (except in the course of computational mathematics) they are not used anywhere. And our graduates, in my opinion, hardly ever will develop their own methods of numerical analysis.
As I understand it, the majority of capable graduates of the OIHVSU of the RSUH take a job as computer linguists at ABBYY, as they cannot compete with students of the Moscow Medical University and the FUPM MIPT for programmers. The other road is freelancers. In my opinion, only a few people from the almost 15-year-old branch graduation are engaged in programming (especially intelligent systems).
As a positive talk about the scientific areas in which our students participate. There are three of them: ontologies and knowledge representation, computational linguistics and intellectual systems.
The first direction is devoted to the development of the original system of knowledge representation "EZOP" . It is a system for the formation and use of ontology based on category theory and formal grammars. The main development language is Prolog.
The second direction is trying to develop and implement general models of linguistic knowledge representation, and apply them to the tasks of syntactic and surface semantic analysis. The main development language is LISP.
The third direction is the creation of intelligent data analysis systems in sociology, criminology, life sciences and robotics. Under the leadership of V.K. In the early 80s, the Finn group of researchers from the VINITI RAS developed an original data analysis method - the DSM-method. In it, by means of multivalued logics, the ideas of induction by D.S. Mill, rigging by K. Popper, reasoning by analogy and abduction by C.S. Pierce. This synthesis of cognitive procedures led to the logical-combinatorial method of machine learning, where causal relationships are extracted from the training set.
Initially, the method was used in pharmacology, where common fragments of chemical formulas of drugs (pharmacophores) were assumed to be the causes of their biological (medical) action. Then the field of application of the method was extended to sociology, where the general features of the description of the respondent’s personality are the “cause” of their behavior or opinion. Then a forensic task arose, where common signs of writing letters can serve as a pretext for finding out the gender and temperament of the writer. Currently, the method is being actively developed for use in medical research. There are attempts to apply the method for training intelligent robots in the Laboratory of Robotics and Artificial Intelligence of the Polytechnic Museum, which also collaborates with the department.
The interaction of the OIS of the RSUH with VINITI RAS has evolved from scientific cooperation into the educational sphere. The base department established at VINITI RAS provides training in many previously described disciplines. Especially I will highlight the courses “Operating Systems” and “Elements of Parallel Programming”, where students have the opportunity to learn how to program in various operating systems and in the multiprocessor cluster deployed at VINITI RAS.
I understand that these are subjective notes on the teaching of artificial intelligence in the RSUH. I think readers will benefit from a look from inside the students of the department of intellectual systems, which can be found in her blog . Recommend…
Currently, in several universities in Moscow there are departments and departments that teach the specialties “Artificial Intelligence”, “Intellectual Systems”, etc. Among them are the Department “Mathematical Theory of Intellectual Systems” of the Faculty of Mathematics of Moscow State University, Department No. 29 “Control Intellectual Systems” and № 22 "Cybernetics" MEPI, the basic department "Intellectual systems" FUPMa MIPT in the EC of the Russian Academy of Sciences ...
Less well-known is the separation of intellectual systems (in the humanitarian sphere) of the RSUH. The ideological inspirer of the creation of this department and its leader is Dr. Konstantinovich Finn , Professor. By the will of fate, I was lucky to learn from him and work under his leadership at the VINITI RAS and at the department described. The principles underlying the department were freely discussed by V.K. Finn and his staff, so I hope my statement will be close to the original. I will make a reservation that this is my statement, so it will be as I understand it.
')
Now, in connection with the transition of education to the system of "bachelor-master" curriculum will change, so in my presentation I will try to reflect the improvements (and deterioration) introduced in the educational process by this reorganization. In 2008, the OIS of the RSUH won the tender of the Federal Agency for Education of the Russian Federation to create training programs in the specialty “Intellectual systems in the humanitarian sphere”. Therefore, the described principles are supposed to be implemented in all universities of Russia, teaching students in this specialty.
Ideologically, the teaching of IHVS specialty is divided into several blocks of academic disciplines that go through all the training.
The first is applied mathematics:
- mathematical analysis,
- algebra,
- Discrete Math,
- Theory of Probability and Mathematical Statistics,
- Computational Mathematics,
- algebraic methods in computer science,
- theory of relational databases.
The second block is logic:
- mathematical logic
- theory of algorithms
- mathematical linguistics,
- axiomatic systems and elements of model theory.
The third is programming:
- programming in languages ​​C and C ++,
- Operating Systems,
- logic programming
- database programming,
- introduction to Windows programming (COM model),
- Internet programming (Javascript language).
Fourth - intellectual technologies:
- Information Systems,
- intelligent systems
- machine learning.
Of course, there are a number of mandatory disciplines in the humanities cycle.
A large number of disciplines are related to the variable part (they can be chosen by students): genetic algorithms and neural networks, fuzzy sets, theory of random processes, the theory of similarity, parallel programming, ontologies and knowledge representation, etc. Many of these courses at RSUH are taught by involved experts.
Historically there is a large block of linguistic disciplines: morphology, syntax, semantics, lexicography, computational linguistics. Due to the reduction of the load on the student, in the new plans this part will be significantly reduced.
However, the ideality of such a curriculum construction faces real difficulties. I consider the main one to be a lack of understanding by applicants of where they go, what subjects are basic, what they will become after graduation from the university. Another difficulty is the inconsistency of training courses.
I will try to explain it on a personal example. I taught and teach the courses "Axiomatic Systems and Elements of Model Theory", "Machine Learning", "Logic Means of Intelligent Systems".
Since the last course depends on the first one, as well as on the basic course “Mathematical Logic”, which is read by V.K. Finn in the first two semesters, I managed to harmonize the material so as to avoid duplication (by organizing, of course, a reminder of the material). I must say that students having studied the method of analytical tables on the 1st course, most of them can use it on the 5th year when they study “Logical tools ...”.
With the course "Machine learning" the situation is radically different. Now my course (under the influence of my personal scientific interests) is shifting towards the probabilistic theory of learning (ala VN Vapnik). But the course of the theory of probability and statistics does not imply the presentation of the material I need (Chernov's inequality, the method of re-sampling, etc.). Therefore, I have to interrupt the presentation of the material by inserts from probability theory. And still, as I believe, students do not understand. Too much different probabilistic material from the algorithmic. I also observe the fact that students forget about NP-completeness (this is needed to prove that sometimes the learning algorithm does not exist). This is aggravated by the fact that the courses “Theory of Algorithms” and “Theory of Relational DBs” interpret it in different terms. Finally, about the theory of information on Shannon (needed for the presentation of the ID3 algorithm of learning decision trees), students do not know anything at all!
This year I even decided to give a course in mathematical statistics, coordinating it with the course of probability theory. I am afraid, however, that this will help me only in a year, when third-year students will start to learn Machine Learning.
As for the wrong goal-setting and inability to single out the most important things for applicants and students, I can only add that I observe a diligent study of mathematical analysis, which they safely forget about course 5, because (except in the course of computational mathematics) they are not used anywhere. And our graduates, in my opinion, hardly ever will develop their own methods of numerical analysis.
As I understand it, the majority of capable graduates of the OIHVSU of the RSUH take a job as computer linguists at ABBYY, as they cannot compete with students of the Moscow Medical University and the FUPM MIPT for programmers. The other road is freelancers. In my opinion, only a few people from the almost 15-year-old branch graduation are engaged in programming (especially intelligent systems).
As a positive talk about the scientific areas in which our students participate. There are three of them: ontologies and knowledge representation, computational linguistics and intellectual systems.
The first direction is devoted to the development of the original system of knowledge representation "EZOP" . It is a system for the formation and use of ontology based on category theory and formal grammars. The main development language is Prolog.
The second direction is trying to develop and implement general models of linguistic knowledge representation, and apply them to the tasks of syntactic and surface semantic analysis. The main development language is LISP.
The third direction is the creation of intelligent data analysis systems in sociology, criminology, life sciences and robotics. Under the leadership of V.K. In the early 80s, the Finn group of researchers from the VINITI RAS developed an original data analysis method - the DSM-method. In it, by means of multivalued logics, the ideas of induction by D.S. Mill, rigging by K. Popper, reasoning by analogy and abduction by C.S. Pierce. This synthesis of cognitive procedures led to the logical-combinatorial method of machine learning, where causal relationships are extracted from the training set.
Initially, the method was used in pharmacology, where common fragments of chemical formulas of drugs (pharmacophores) were assumed to be the causes of their biological (medical) action. Then the field of application of the method was extended to sociology, where the general features of the description of the respondent’s personality are the “cause” of their behavior or opinion. Then a forensic task arose, where common signs of writing letters can serve as a pretext for finding out the gender and temperament of the writer. Currently, the method is being actively developed for use in medical research. There are attempts to apply the method for training intelligent robots in the Laboratory of Robotics and Artificial Intelligence of the Polytechnic Museum, which also collaborates with the department.
The interaction of the OIS of the RSUH with VINITI RAS has evolved from scientific cooperation into the educational sphere. The base department established at VINITI RAS provides training in many previously described disciplines. Especially I will highlight the courses “Operating Systems” and “Elements of Parallel Programming”, where students have the opportunity to learn how to program in various operating systems and in the multiprocessor cluster deployed at VINITI RAS.
I understand that these are subjective notes on the teaching of artificial intelligence in the RSUH. I think readers will benefit from a look from inside the students of the department of intellectual systems, which can be found in her blog . Recommend…
Source: https://habr.com/ru/post/93183/
All Articles