0x010088F6312065A70A44B630A6848498EAD80093CFEAFE161DC94D83C7C5C9A5A748B8

The course "Cognitive Architectures" undergoes continuous development every year, with the aim of adapting its contents to new technological advancements and ensuring ongoing improvements in the teaching approach. In the fall term of 2023, the course was conducted through live lectures held in lecture hall R5, which were recorded and made accessible to students unable to attend in person. Lecture slides were uploaded to Blackboard shortly before each session. From the instructor's perspective, live lectures in a physical hall still offer advantages over digital ones, allowing for more immediate student feedback with lower barriers. However, it was noted that Norwegian students, despite encouragement, tend to be somewhat hesitant in taking the initiative to interrupt and ask questions. Efforts were made to enhance the course slides, addressing issues present in previous sets. Regarding content, the initial three lectures focused on explicating the concept of "cognition," particularly in relation to technical systems claiming to exhibit "intelligence." The goal was to foster a critical mindset among students towards exaggerated claims of intelligence in such systems. The course incorporated an extensive reading list, comprising both mandatory and optional materials, consisting of freely available papers and articles to eliminate the need for additional textbook purchases. This self-study component required several hours of weekly reading to reinforce concepts introduced in lectures. Motivating students to consistently engage with course materials, make up for missed sessions, and overcome temporary disruptions in their learning progress is a common challenge. To address this, and to significantly enhance students' chances of passing the final exam in December, homework assignments were introduced, usually with a 2-week deadline for completion. Although there were initial challenges in promptly grading the first homework assignment, adjustments were made to subsequent assignments to maintain the 2-week timeframe for preparation and submission. An extra homework assignment was introduced to accommodate students facing difficulties due to schedule deviations, ensuring ample chances for exam admission, while keeping the number of required approvals the same. Weekly tutorial hours, scheduled immediately after lectures, provided opportunities for students to ask questions related to lecture content and homework assignments.

While providing recorded lectures throughout the entire semester is advantageous, it also poses the risk of students delaying their individual post-processing of lectures. To address this, there is consideration of limiting access to recorded lectures to one or two weeks after delivery, emphasizing that last-minute "binge-watching" is not a successful strategy for exam preparation.

As already indicated, direct feedback from students on the contents of the course and their perception of the relevance of the course contents is usually only coming from the most motivated students. Therefore, it should be taken "cum grano salis" what can be reported here: I had mostly positive feedback from those students who consider also the somewhat "philosophical" introduction into the subject matter of cognition as stimulating. As the course will be elective from fall term 2024 and will be postponed in the curriculum by 1 year, I intend to introduce more contents from cognitive science, and reduce on some fundamentals of classical reasoning and machine learning that were necessary with the position of the course in the curriculum, since these foundations will be covered in other courses.

NTNU:TIA:REST

Institutt for datateknologi og informatikk

Department of Computer Science

Ingen trykte eller håndskrevne hjelpemidler tillatt. Bestemt, enkel kalkulator tillatt.

Engelsk

Mer om vurdering

Ved utsatt eksamen (kontinuasjonseksamen) kan skriftlig eksamen bli endret til muntlig.Dersom det etter utsatt eksamen fremdeles er karakteren stryk, eller ved gjentak, kan obligatorisk aktivitet fra tidligere semester godkjennes av instituttet.Eksamen blir kun gitt på engelsk, men studentene kan svare på norsk om de foretrekker det.

Faglig innhold

Kognitive arkitekturer bidrar til økt integrert forståelse om oppbyggingen av intelligente systemer og hvordan disse interakterer med den eksterne verden. Emnet tar opp underliggende infrastrukturer og hvordan forskjellige subsystemer samspillerer for å skape en koherent kognisjon. Det fokuseres blant annet på korttids- og langtids-hukommelse, representasjon av mentale strukturer, funksjonelle prosesser, og læringsmekansimer. I denne kontekst ser en også på evaluering av brukergrensesnitt som kognitiv kompleksitetsanalyse, og på avansert multimodal interaksjon.

Kursmateriell

Oppgis ved semesterstart (samling av artikler)

Læringsutbytte

Kunnskaper: Emnet skal gi studenten en generell innføring i begrepsapparat, metoder og teknikker innen kognitive arkitekturer. Ferdigheter: Studenten skal være i stand til å benytte aktuelle metoder og teknikker.Generell kompetanse: Studenten skal skal gjennom praktiske og teoretiske oppgaver forstå og anvende kognitive arkitekturer og ha et helhetlig bilde av intelligente systemer. Kandidaten skal kjenne fagfeltets historikk og grunnlag hentet fra kunstig intelligens, kognitive vitenskaper og bla menneske-maskin interaksjon.

Læringsformer og aktiviteter

Forelesninger, kollokvier, selvstudium og øvinger. Et antall obligatoriske øvinger må bli godkjent for å kunne gå opp til eksamen.

Øvinger

Kognitive arkitekturer

Datateknikk og informasjonsvitenskap

Informasjonsteknologi og informatikk

Informatikk - bachelorstudium

Datateknologi - masterstudium (5-årig)

Web/Lists(guid'8b698a2d-6ad0-4f7d-b3be-219cc1979f02')/Items(26490)/FieldValuesAsText

The course "Cognitive Architectures" is under continuous development from year to year, with the intention to adapt the course contents to new developments on the technological side, but of course also with respect to continuous improvement on the didactic side.In the fall term 2022, the course was based on live lectures in lecture hall H3, recorded and also made available to those students that could not attend the course in person. The experience from the teacher's side was that live lectures in an actual lecture hall provide still advantages compared to digital ones, as there is a chance for feedback from the students (more precisely: the barriers for feedback are lower). On the other hand, I experienced again that the Norwegian students are somewhat "shy" to take initiative to interrupt and ask questions, even though they were encouraged to do so when necessary. The first 2 lessons and the guest lecture could not be recorded due to technical problems, but audio recordings were made available.The course slides have been improved, eliminating some flaws from previous slide sets.Concerning the contents, more weight was put on explaining in the first two lectures what "cognition" actually is, in particular with a focus on technical systems that are claimed to behave "intelligently". It was tried to strengthen the critical attitude of the students towards excessive claims on "intelligence" for such systems.We offered weekly tutorial hours immediately after the lectures. What was unpleasant both for the lecturer as well as the students is that we had to move to another locality at the other end of the campus. The tutorial hours are occasions to ask any question related to the lecture contents and/or the homework assignments.Due to Covid sickness of the lecturer, a delay of initially two weeks occurred in the early half of the course, not in the lectures (which were held digital 2 times at that time), but with respect to the assignments, which caused some compression of the available time period for solving the later assignments towards the end of the lecturing period.

A typical problem with motivating the students to continuously follow the lectures, catch up with missed teaching units, or other temporary disruptions of their learning progress is to see that the students steadily work with the teaching materials made available. For this reason, but also in order to significantly increase the chances of the students to pass the final exam (typically in December), we provide homework assignments, usually scheduled in 2 weeks periods. 4 of the 5 homework assignments need to be approved. In the present period, we even provided a 6th homework block with old exam questions for selfstudy (not graded), and due to some delays in the homework publication schedule (due to the mentioned sickness period of the course responsible) we lowered for this time only the requirement to 3 approved homeworks. Despites this, we experienced a substantial amount of individual problems with students who struggled to get these homeworks approved.The fact that we offer the lectures in recorded form throughout the complete semester is certainly a good thing, but it increases also the risk that the students postpone their individual lecture post-processing. We are considering making the lectures available only for one or two weeks after they were held, in order make it clear that "binge-watching" the lectures in the week before the exam is no good success strategy.

There is a significant discrepancy between my (emneansvarig) intention to provide fair and individual feedback on the students' performance, in particular the exam grading, and the given processes. Particularly the present course (Cognitive Architectures) comes with the intention to train and extend the students' capability to compare and critically evaluate different, partially strictly contrasting approaches in the literature and in current research, This is reflected in the fact that the written exam at the end of the course contains a large portion of questions and tasks that are to be answered by mini-essays (some few sentences). To evaluation to what extent these answers actually are associated to the contents of the lecture, which contain yearly update materials, is difficult tor grading persons who did not attend the lectures and did not read the quite substantial amount of reading materials which are mostly papers and encyclopaedic articles. This is -- obviously -- different from many other courses where competences are tested that should be available to any academic familiar to the regarded domain. I tried also to move away as far as possible from testing the familiarity of the students with catalog-structured knowledge, to be tested in Y/N questions or multiple choice tests. However, going away from these strictly formalised test methods which require the students only to blindly learn and memorise catalogs of atomic knowledge elements immediately leads to a steep increase of grading effort, thus conflicts with the constraints given for the exam process.

Norsk

Ved utsatt eksamen (kontinuasjonseksamen) kan skriftlig eksamen bli endret til muntlig.

Web/Lists(guid'8b698a2d-6ad0-4f7d-b3be-219cc1979f02')/Items(20994)/FieldValuesAsText

In fall 2020, the course has been held for the first time by Rudolf Mester (RM). After the experiences with running this course for the first time, RM decided to stepwise lead the course to a higher percentage of material from recent scientific developments. The most important adaption in that respect is to provide an insight in basic principles of machine learning ("connectionist approaches", in the terminology of the course) and statistical (most Bayesian reasoning. The course cannot and will not provide a broad and detailed introduction of these areas of AI, but will introduce selected principles, and will try to put the "classical" cognitivistic (symbol-based) approaches in a comparative context with the recent ML approaches. To work out the details of the ML-based approaches is a task for which a set of dedicated courses at IDI already exist. In the fall term 2021, the presentation of statistical reasoning (in particularly Bayesian reasoning) as well as the presentation of connectionist approaches (from perceptron to deep learning) has been intensified. Furthermore, the details of the fundamental components of Cognitive Systems has been presented in more detail, oriented on the very valuable survey paper by Kotseruba and Tsotsos.

GenereltThe current course responsible (Rudolf Mester) took over the course from Agnar Aamodt, a colleague who retired in summer 2020.In previous passes, the course has been given in regular physical presence mode.Due to the Corona measures taken in 2020, the course had to be restructured towards a mainly digital format, meaning that the lectures have been given in digital form in fall 2020, and have been again given in that format in fall 2021. The presentations / lectures have of course been given live, but have also been recorded and made available to the students throughout the complete term.The contents of the course are continuously steered towards the principle of including a reasonable mixture of “classical” contents from the field of AI and Cognitive Architectures (e.g. existing architectures like SOAR, ACT-R, ICARUS…) and presenting also methods and technologies which have more recently been developed. It is also one aim followed with the presentation of the material to illustrate the huge variability of reasoning methods, ranging from deliberative approaches including extensive planning to reactive approaches.The course was finalized by an extra lecture (2 hrs) offering a quick retrospective about what has been covered in the course.For the course, a good set of lecture slides was available from previous passes. However, this lecture slide set has been updated with topics of recent interest, and a number of new aspects have been added to the course contents, eg. some more elements on Deep Neural Networks, some aspects of "Attention", and some aspects of contemporary AI systems that use and need cognitive processes, for instance autonomous driving and robotics. The motivation for these changes was to keep the course up to date, and show also current application perspectives of the course contents to the students Parallel to the lectures, 5 assignment sheets have been provided in the course of the semester, where the students' solutions to 4 of these had to be approved in order to be admitted to the final exam. These assignment sheets and the students' question related to these were discussed in a weekly tutorial session (as mentioned). The contents of the Cognitive Architectures course are different and widely complementary to the Introduction to AI course ongoing in parallel. The course exercise hours / tutorial hours have been offered as weekly events where the main lecturer plus one or two teaching assistants have been present in a Blackboard Collaborate event. During these hours, the students had the opportunity to ask questions related to the current homework sheets, or general questions related to the lecture or the general field covered by the course.”. LæringsutbytteAs already mentioned above, with the transition of course responsible, the course contents have been adapted to newest trends in the field of AI, reflecting the still increasing impact of Deep Learning on the AI field, but also stressing that "Symbolic AI" still plays a considerable role. It is the explicit intention of the course responsible (RM) to show the demand of explainability for AI and Cognitive Systems. This demand for explainability supports the principle that cognition is (much) more that association making. The width of techniques and methods and theories that have been developed over decades 'crystallised" in existing Cognitive Architectures, such as SOAR, ICARUS, ACT-R, but also in "emergence" approaches. The lectures typically focussed on one or two exemplary cognitive architectures, or alternatively on an overview of methodological approaches, such as Bayesian reasoning, Fuzzy Inference, or Artificial Neural Networks. The course included several guest lectures from leading researchers in the field, such as John Tsotsos (on attention) and Christoph von der Malsburg (on neural architectures).The course content is presented to the students in terms of lectures which are given on the basis of slide sets presented in live lectures, and recorded in Blackboard (BB) for later study and exam preparation purposes. UndervisningsformerThe Cognitive Architectures course is designed in such a way that students are expected to continuously attend (or at least post-watch) the weekly lectures, re-read the provided lecture slides and the additional reading material after each lecture, and probe their understanding of the material by solving the tasks given in the assignment sheets. Questions that arise during the study of this material are dealt with in the weekly tutorial sessions. Due to the digital format, the interaction with students was provided by the chat channels during the lectures, to the post-lecture exercise/tutorial hours, and the possibility to ask questions via email.The course responsible (RM) had the impression that the speed and detail depth of the lectures was complying with the students requirements and expectations, at least this was the conclusion from asking students directly during the tutorial hours.

The course responsible (RM) had the impression that providing lecture recordings throughout the whole term resulted in many students using these lectures only very late, shortly before the exam. It is considered to limit the availability of the recordings to 1 or 2 weeks after the lecture, and a final phase of several weeks before the exam.

Kognitive arkitekturer bidrar til økt integrert forståelse om oppbyggingen av intelligente systemer og hvordan disse interakterer med den eksterne verden. Emnet tar opp underliggende infrastrukturer og hvordan forskjellige subsystemer samspillerer for å skape en koherent kognisjon. Det fokuseres blant annet på korttids- og langtids-hukommelse, representasjon av mentale strukturer, funksjonelle prosesser, og læringsmekansimer. I denne kontekst ser en også på evaluering av brukergrensesnitt som kognitiv kompleksitetsanalyse, og på avansert multimodal interaksjon. 

Forelesninger, kollokvier, selvstudium og øvinger. Et antall obligatoriske øvinger må bli godkjent for å kunne gå opp til eksamen. 

Kunnskaper: Emnet skal gi studenten en generell innføring i begrepsapparat, metoder og teknikker innen kognitive arkitekturer. Ferdigheter: Studenten skal være i stand til å benytte aktuelle metoder og teknikker. Generell kompetanse: Studenten skal skal gjennom praktiske og teoretiske oppgaver forstå og anvende kognitive arkitekturer og ha et helhetlig bilde av intelligente systemer. Kandidaten skal kjenne fagfeltets historikk og grunnlag hentet fra kunstig intelligens, kognitive vitenskaper og bla menneske-maskin interaksjon. 

Web/Lists(guid'8b698a2d-6ad0-4f7d-b3be-219cc1979f02')/Items(15454)/FieldValuesAsText

The current course responsible (Rudolf Mester) took over the course from Agnar Aamodt, a colleague who retired in summer 2020.In previous passes, the course has been given in regular physical presence mode.With the Corona measures in place, the course had to be restructured towards a mainly digital format, meaning that the lectures had to be given in digital form,and also the course exercise hours / tutorial hours had to be offered as "semi-physical". This means that the tutorial hours (usually on the same day as the lecture) were held in a lecture hall in Realfagsbygget, making them accessible also from remote. With the unclear and at times varying Corona situation, most students chose to participate in these tutorial hours digitally.For the course, a good set of lecture slides was available from previous passes. However, this lecture slide set has been updated with topics of recent interest, and a number of new aspects have been added to the course contents, eg. some more elements on Deep Neural Networks, some aspects of "Attention", and some aspects of contemporary AI systems that use and need cognitive processes, for instance autonomous driving and robotics. The motivation for these changes was to keep the course up to date, and show also current application perspectives of the course contents to the students.

As already mentioned above, with the transition of course responsible, the course contents have been adapted to newest trends in the field of AI, reflecting the still increasing impact of Deep Learning on the AI field, but also stressing that "Symbolic AI" still plays a considerable role. It is the explicit intention of the course responsible (RM) to show the demand of explainability for AI and Cognitive Systems. This demand for explainability supports the principle that cognition is (much) more that association making. The width of techniques and methods and theories that have been developed over decades 'crystallised" in existing Cognitive Architectures, such as SOAR, ICARUS, ACT-R, but also in "emergence" approaches. The lectures typically focussed on one or two exemplary cognitive architectures, or alternatively on an overview of methodological approaches, such as Bayesian reasoning, Fuzzy Inference, or Artificial Neural Networks. The course included several guest lectures from leading researchers in the field, such as John Tsotsos (on attention) and Christoph von der Malsburg (on neural architectures).The course content is presented to the students in terms of lectures which are given on the basis of slide sets presented in live lectures, and recorded in Blackboard (BB) for later study and exam preparation purposes. Furthermore, 5 assignment sheets have been given out in the course of the semester, where the students' solutions to 4 of these had to be approved in order to be admitted to the final exam. These assignment sheets and the students' question related to these were discussed in a weekly tutorial session (as mentioned). The contents of the Cognitive Architectures course are different and widely complementary to the Introduction to AI course ongoing in parallel.

The Cognitive Architectures course is designed in such a way that students are expected to continuously attend (or at least post-watch) the weekly lectures, re-read the provided lecture slides and the additional reading material after each lecture, and probe their understanding of the material by solving the tasks given in the assignment sheets. Questions that arise during the study of this material are dealt with in the weekly tutorial sessions. As there was (in H2020) no physical presence during the lectures, the interaction with students was limited to the chat channels during the lectures.The course responsible (RM) had the impression that the speed and detail depth of the lectures was complying with the students requirements and expectations, at least this was the conclusion from asking students directly during the tutorial hours.

Kognitive arkitekturer bidrar til økt integrert forståelse om oppbyggingen av intelligente systemer og hvordan disse interakterer med den eksterne verden.
Emnet tar opp underliggende infrastrukturer og hvordan forskjellige subsystemer samspillerer for å skape en koherent kognisjon.  Det fokuseres blant annet på korttids- og langtids-hukommelse, representasjon av mentale strukturer, funksjonelle prosesser, og læringsmekansimer.  I denne kontekst ser en også på evaluering av brukergrensesnitt som kognitiv kompleksitetsanalyse, og på avansert multimodal interaksjon.

Kunnskaper: Emnet skal gi studenten en generell innføring i begrepsapparat, metoder og teknikker innen kognitive arkitekturer. 
Ferdigheter: Studenten skal være i stand til å benytte aktuelle metoder og teknikker.
Generell kompetanse: Studenten skal skal gjennom praktiske og teoretiske oppgaver forstå og anvende kognitive arkitekturer og ha et helhetlig bilde av intelligente systemer. Kandidaten skal kjenne fagfeltets historikk og grunnlag hentet fra kunstig intelligens, kognitive vitenskaper og bla menneske-maskin interaksjon.

Web/Lists(guid'8b698a2d-6ad0-4f7d-b3be-219cc1979f02')/Items(9769)/FieldValuesAsText

Kognitive arkitekturer - TDT4137