0x010088F6312065A70A44B630A6848498EAD80093CFEAFE161DC94D83C7C5C9A5A748B8

This course is in continuous development, because new topics in statistics are emerging all the time - and new developments in student active learning and technology, etc. The last run of the course was in 2021, and then the feedback was that the course had too many topics and there was a suggestion to remove the part on neural network. This was done in 2023, and the neural network part was replaced by a more general first part on core concepts in statistics (that have not to a large enough extent been covered in other statistics courses at IMF). The main topics covered were in 2023<ul><li>Part 1: Core concepts [new: missing data, including multiple imputation]</li><li>Part 2: Shrinkage and regularization [mainly unchanged from 2021]</li><li>Part 3: Ensembles (of trees and other methods) [mainly unchanged from 2021]</li><li>Part 4: Explainable AI [unchanged from 2021]</li><li>Part 5: Closing</li></ul> In 2021 we still had issues with the pandemic, and the teaching was 3 hours (mainly digitally) each week of the semester - with the use of breakout-rooms for group work during the lectures. Feedback from the students (2021) was that the teaching was done at a too fast pace. In 2023 I gave 2 hours lectures 2 times each week. In the lectures the students were placed in groups, and discussed topics integrated in the lectures. The group discussions were a popular part of the course, according to the students (student evaluation). In 2021 evaluation of the course were a portfolio (with three group elements) and an oral exam (both at pass/fail). This was in 2023 changed to obligatory work (two elements: data analysis project and article presentation, both in groups) and the oral exam (pass/fail). This means that only one data analysis project was present in 2023. In addition, some effort was made to make the article presentation more inline with the course topics than in 2021 (where a broader range of topics were presented). As in 2021 a group contract was made for each group, and for the data analysis a review group was assigned for each group for the first submission (the second submission was final). For the article presentation each group presented and also were dicussant group for other presentations. New of 2023 was that the statistics phd-students acted as chairs for some of the article presentations, making the presentations more similar to a conference presentation.

See the feedback from the students (IEs student survey). 15 students met the compulsory requirements to present themselves at the exam. 13 students signed up for the oral exam. A re-take exam was arranged in the autumn of 2023.

MA8701 is a course on the PhD-level in statistics. Therefore the background requirements for the course are listed to be core statistics courses that MA8701 build upon. In particular the master level course of TMA4315 Generalized linear models is important for several parts of the MA8701-course, and the TMA4268 Statistical learning course. However, many students that sign up for the course have not taken the GLM course, and in particular they also have limited knowledge of classic statistical inference (confidence intervals, hypothesis tests, multiple testing). This is a problem in the way that the level of understanding for the statistical inference part of the course (which is central) is lower among the students than if the course requirements were kept. See the reference group report, where the group suggests to remove some of these prerequisite courses. I do not agree with that, and find that the GLM course should be taken before MA8701. However, the prerequisites are not formally binding.

NTNU:TIA:REST

Institutt for matematiske fag

Department of Mathematical Sciences

Spesifiserte trykte og håndskrevne hjelpemidler tillatt. Ingen kalkulator tillatt.

Engelsk

Norsk

Forkunnskapskrav

TMA4267 Lineære statistiske modeller, TMA4295 Statistisk inferens, TMA4300 Beregningskrevende statistiske metoder, TMA4268 Statistisk læring eller tilsvarende kunnskaper, TMA4315 Generalized linear models. God forståelse og erfaring med R, eller Python, for statistisk dataanalyse.

Link til informasjon om emnet

Faglig innhold

Emnet har fokus på sentrale begreper og metoder innen statistisk inferens og læring. Innholdet er en utvidelse av hva som tidligere er lært i emnene som listet under Forkunnskapskrav.

Kursmateriell

The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Second Edition (Springer Series in Statistics, 2009) by Trevor Hastie, Robert Tibshirani, and Jerome Friedman. (Siste oppdaterte versjon finnes fra https://web.stanford.edu/~hastie/ElemStatLearn/.) I tillegg vil det bli bruke utvalgt annet materiale (kapitler fra bøker og journalartikler). Detaljert informasjon blir gitt ved semesterstart.

Læringsutbytte

1. Kunnskap. Forstå og kunne gjøre rede for sentrale teoretiske aspekter i statistisk inferens og læring. Forstå og kunne gjøre rede for hvordan man bruker metodene fra statistisk inferens og læring til å utfører en god dataanalyse. Være i stand til å evaluere styrker og svakheter ved metodene og velge mellom dem i en gitt dataanalysesituasjon. 2. Ferdigheter. Kunne analysere et datasett med metoder fra statistisk inferens og læring i praksis (ved hjelp av R eller Python), og kunne gi en god presentasjon og diskusjon av valgene som er gjort og resultatene som er funnet. 3. Kompetanse. Studentene vil være i stand til å delta i vitenskapelige diskusjoner, og sette seg inn i forskning presentert i statistiske tidsskrifter. De vil være i stand til å delta i anvendte prosjekter der data skal analyseres med metoder innen statistisk inferens og læring.

Læringsformer og aktiviteter

Forelesninger, evt. som ledet selvstudium. Praktisk prosjekt i dataanalyse (anvendelse av teorien i emnet med bruk av R eller Python) med muntlig presentasjon (bestått/ikke-bestått). Emnet foreleses normalt hvert annet år, og bare dersom et tilstrekkelig antall interesserte melder seg. Emnet foreleses våren 2023. Dersom det melder seg få studenter, vil kurset gis som ledet selvstudium.

Arbeider

Avanserte statistiske metoder innen inferens og læring

Statistikk

Matematiske fag - doktorgrad (ph.d.)

Web/Lists(guid'8b698a2d-6ad0-4f7d-b3be-219cc1979f02')/Items(22683)/FieldValuesAsText

This is a course designed for Ph.D students in statistics who want to get an updated theoretical understanding of current models and methods used within statistical learning, with focus on the statistical inference part of statistical learning. However, the course is also of interest among other Ph.D. students (engineering disciplines and mathematics) and master students. This makes it hard to give all students teaching resources at their optimal level, this is, in spite of the clear prerequisites defined for the course. The theoretical and learning elements of the course is presented in <a href="https://wiki.math.ntnu.no/_media/ma8701/2021v/overviewall.png?w=800&amp;tok=c4db79" target="_blank">this drawing</a>. I assess the quality of the learning material to be high, but the quality of the lectures can be improved. The quality of the group work appears very high. In total I assess the quality of the course to be good. Suggested changes to the course curriculum: This is a course that aims to be up to date with the research frontier in statistical learning and inference, and when the course run the next time in V2023 then the field of XAI hopefully is even stronger and the course can get deeper into this. This also applies to the selective inference (Part 1, L4), hyperparameter tuning and inference for comparing prediction methods (Part 2) and Bayesian nets (Part 3). However, as also commented on by the reference group, we might drop one of the four parts to be able to focus more on the other parts and to give the lectures a lower pace. Tentatively the part to remove can be Part 3: Neural nets. Also, the role of the articles for the article presentation task may be stronger linked to the curriculum and maybe also part of the reading list.

Communication:The learning outcomes were<ul><li>presented on the course wiki-page</li><li>presented in the first lecture</li><li>discussed in connection with the portfolio (see Evaluation forms)</li><li>discussed (are they fulfilled) in the last lecture</li></ul> Learning outcome up to date?1. Knowledge. Understand and explain the central theoretical aspects in statistical inference and learning. Understand and explain how to use methods from statistical inference and learning to perform a sound data analysis. Be able to evaluate strengths and weaknesses for the methods and choose between different methods in a given data analysis situation. The knowledge outcome is up to date. 2. Skills Be able to analyse a dataset using methods from statistical inference and learning in practice (using R or Python), and give a good presentation and discussion of the choices done and the results found. The skills outcome is up to date. 3. Competence The students will be able to participate in scientific discussions, read research presented in statistical journals, and carry out research in statistics at high international level. They will be able to participate in applied projects, and analyses data with methods from statistical inference and learning. The competence outcome was aimed at Ph.d. students in statistics - especially the "carry out research in statistics" is not fulfilled for the non-statistics students. The following change to the Competence outcome will be made before the next time the course is given: 3. Competence The students will be able to participate in scientific discussions, read research presented in statistical and machine learning journals, and carry out statistical analysis to be published in high quality journals. They will be able to participate in applied projects, and to analyse data using methods from statistical inference and learning. Connection to other courses:Prerequisite knowledge is: TMA4267 Linear Statistical Methods, TMA4295 Statistical inference, TMA4300 Computer intensive statistical methods, TMA4268 Statistical learning or equivalent knowledge. Good understanding and experience with R, or with Python, for statistical data analysis. Here the course TMA4315 Generalized linear models is missing, and will be added. These prerequisite courses had mostly focus on knowledge, but (apart from TMA4295) also had a skills component connected to data analysis. Many of the participants in MA8701 V2021 did not have this prerequisite knowledge, which made it hard to understand all elements of the course in depth. However, at the final oral exam the necessary depth was demonstrated by all students. The learning outcomes in MA8701 is aligned with the outcomes in MA8702 which is a second course in computational methods - built with the same principles as MA8701.

Learning activities: lectures, exercises, office hours, group work, data analysis, article presentation.Lectures are mainly aimed towards the knowledge outcome, while group work indirectly target the knowledge outcome by working with the skills and competence outcome. Main information channel was the course wiki page at: https://wiki.math.ntnu.no/ma8701/2021v/startCommunication of information from lecturer has mainly been by email. Lectures: Due to the corona situation only 3 (L8+9+14) out of the 14 lectures were hybrid, the remaining 11 were digital. Slides were available before and after the lectures. There were 5 guest lectures, and all were recorded. All lectures lasted 3 hours. The lectures had 90-100% student attendance. Students asked questions in the chat and orally. Break-out rooms were used for student discussions, but in retrospect should have been used more. Polls in zoom were also used. In total, the lecture part of the course could have been more interactive and also at a lower speed (little time spent on each topic).Office hours: All office hours were on zoom, with a hybrid option for parts of the semester. The office hours were well visited in the start of the course, with decreasing attendance towards the end of the course. First the focus was on exercises, then on the projects and finally the oral exam. This activity mainly addressed the knowledge outcome. Group work: The lecturer met with all the 9 nine groups in the course (on zoom), and communicated with them with on throughout the semester. Each group set up their own choice of collaboration channel (in person, zoom, messenger, teams, ...) and 8/9 groups used github for code sharing and collaboration. A group contract was used for the second data analysis project. I suggest to do this for the first project in V2023. Padlets were used to share information on the class level on packages in R and Python and on where to find good data sets. The motivation of the students was high - with high attendance at the lectures, and in the data analysis projects and the article presentation much work was done. The learning environment have been including, and the communication between students and student-lecturer has reflected mutual respect and trust.

The evaluation of the course is pass/fail where the pass-limit is at 70%. The evaluation consisted in V2021 of a portfolio (75%) and an oral exam (25%). (This is a corona change from the original 30% portfolio and 70% oral - starting in V2019, and before V2019 this was 100% oral.) The portfolio consisted of three elements, where all were performed as group work: two written data analysis projects- one with peer review and submission after the review, and one article presentation with discussants (on zoom or hybrid in a zoom room=S21) The portfolio directly assess the skills and competence outcome - and indirectly the knowledge outcome. The oral exam mainly evaluate the knowledge outcome. The oral (individual) exams were held during four days of May with 30 minutes slots. The lecturer and one internal sensor were present in addition to the student. 14 students met in person (room 822 worked very well with blackboard and TV for showing plots and questions) and the remaining 7 on zoom. 23 students passed the portfolio and 21 met and passed the oral exam.

Alle trykte og skrevne

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

TMA4267 Lineære statistiske modeller, TMA4295 Statistisk inferens, TMA4300 Beregningskrevende statistiske metoder, TMA4268 Statistisk læring eller tilsvarende kunnskaper. God forståelse og erfaring med R, eller Python, for statistisk dataanalyse.

Emnet foreleses normalt hvert annet år, og bare dersom et tilstrekkelig antall interesserte melder seg. Foreleses neste gang våren 2021. Dersom det melder seg få studenter, vil kurset gis som ledet selvstudium.

Emnet har fokus på sentrale begreper og metoder innen statistisk inferens og læring. Sammen med emnene "MA8704 Sannsynlighetsteori og asymptotiske teknikker" og "MA8702 Videregående beregningskrevende statistiske metoder" danner det en teoretisk basis for doktorgradsstudenter innen statistikk.

Innholdet er en utvidelse av hva som tidligere er lært i emnene som listet under Forkunnskapskrav. 

The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Second Edition (Springer Series in Statistics, 2009) by Trevor Hastie, Robert Tibshirani, and Jerome Friedman. (Siste oppdaterte versjon finnes fra https://web.stanford.edu/~hastie/ElemStatLearn/.) I tillegg vil det bli bruke utvalgt annet materiale (kapitler fra bøker eller journalartikler). Detaljert informasjon blir gitt ved semesterstart.

Forelesninger, evt. som ledet selvstudium. Praktisk prosjekt i dataanalyse (anvendelse av teorien i emnet med bruk av R eller Python) med muntlig presentasjon (bestått/ikke-bestått).

1. Kunnskap.
Forstå og kunne gjøre rede for sentrale teoretiske aspekter i statistisk inferens og læring. Forstå og kunne gjøre rede for hvordan man bruker metodene fra statistisk inferens og læring til å utfører en god dataanalyse. Være i stand til å evaluere styrker og svakheter ved metodene og velge mellom dem i en gitt dataanalysesituasjon.

2. Ferdigheter.
Kunne analysere et datasett med metoder fra statistisk inferens og læring i praksis (ved hjelp av R eller Python), og kunne gi en god presentasjon og diskusjon av valgene som er gjort og resultatene som er funnet.

3. Kompetanse.
Studentene vil være i stand til å delta i vitenskapelige diskusjoner, sette seg inn i forskning presentert i statistiske tidsskrifter, og utføre forskning innen statistikk på høyt internasjonalt nivå. De vil være i stand til å delta i anvendte prosjekter der data skal analyseres med metoder innen statistisk inferens og læring.

Web/Lists(guid'8b698a2d-6ad0-4f7d-b3be-219cc1979f02')/Items(11615)/FieldValuesAsText

Avanserte statistiske metoder innen inferens og læring - MA8701