teaching schedule, now ready

Follow-up to Machine Learning – a hands–on course from Rudo's blog

Dear all,

the schedule of the course has now become clearer: we continue to teach every Monday 0930-1230 in room E4.28 (except for one day in January and one day in February) for 10 weeks overall. For the 1st 5 weeks (with 4 more weeks to come), I will actually "teach", i.e. with prepared lectures for 2 hours 0930-1130, while Djenabou will be available for 1 hr more 1130-1230 to help you with the exercises. Then in the last 5 weeks, Djenabou will be available 0930-1230 to help with your project work (she won't do it for you, but might be able to help with technical problems).

A full schedule should be available soon.

See you next Monday!

-Rudo

Rudo Roemer : 12 Dec 2019 15:53 |  Tags: Fastai Keras Machine-Learning Tensorflow |  Comments (0) |  Close comments |  Report a problem

Thoughts on the 1st lecture session

Writing about web page /rroemer/entry/installing_your_own/

Writing about an entry you don't have permission to view

Dear all,

[1] today in the lecture, we saw that it was not entirely simple to install your own ML machine with Tensorflow/Keras and FastAI. I have now made a blog entry on how to do this. You can find it here.

[2] In addition, when running the FastAI code, we saw that it might take a long time even for just the pets categorization. I just ran the pets notebook (please download a new version) and have made the pets-stage-1 and pets-stage-1-50 trained models available on the download area. You need to move these models into where FastAI keeps them which in my set-up is

/home/phsht/.fastai/data/oxford-iiit-pet/images/models

Just copy the twp .pth files there and you should be able to run the pets code after the

learn.load('pets-stage-1')

fastai commands in the notebook.

[3] When you "play" with the notebook called "FastAI-lesson2-download.ipynb", note that the file data.zip contains the cats/teddys/wolves that I managed to download and use for the exercise.

[4] Some of you argued that your own machines may not be able to do meaningful ML as you do not have a GPU. Have a look at what the FastAI guys have collected in terms of possible free GPU instances, i.e. Using a GPU. I have not yet used any of these services, but would be keen to learn from your experiences!

OK, keep checking this blog for up-to-date information before we meet again next week.

Rudo Roemer : 09 Dec 2019 16:29 |  Tags: Fastai Keras Machine-Learning Tensorflow |  Comments (2) |  Close comments |  Report a problem

Things to do before the 1st lectures

Follow-up to Machine Learning – a hands–on course from Rudo's blog

Dear all,

[1] some of you will have registered for the short Machine Learning (ML) course that we are planning to give, starting next week. ML is best "learned" by doing it, so we are planning to use the computer room E4.28 for exercises during all the lectures. But we think you might also find it more useful if you were to bring your laptops along and install+run the ML codes there.

[2] We are expecting to run all the ML on a Ubuntu installation (Francois Chollet, one of the key people in the field, says "it’s possible to use Windows, too, but I don’t recommend it"). Other Linux installations should also work. If you only have Windows, Rudo recommends installing

VirtualBox, https://www.virtualbox.org/

with a Ubuntu 19.10 installation (http://releases.ubuntu.com/19.10/).

[3] During the course, we intend to offer you downloadable ML source codes via Rudo's blog at

https://blogs.warwick.ac.uk/rroemer/entry/machine_learning_a/

where you can also see a copy of this email. The direct download link is in the first entry of the blog and given here as well, i.e.

https://cloud.csc.warwick.ac.uk/index.php/s/JomfpoMj87X3Jbk

[4] We want to share literature on ML with you using Mendeley, i.e. https://www.mendeley.com/. Although the Mendeley Desktop might be convenient to install (we have it on our machines), the online interface could be ok for the course. But in order to "share" PDFs with you, you need to have a (free) Mendeley account. Please create one.

[5] Last, we, Djena and Rudo, are both new to ML, so this will be a course where we will all explore ML together! We hope you'll enjoy it.

- Rudo and Djena

Rudo Roemer : 06 Dec 2019 09:13 |  Tags: Fastai Keras Machine-Learning Tensorflow |  Comments (0) |  Close comments |  Report a problem

Machine Learning – a hands–on course

Dear all,

for this course, you should be able to find all course material for download at

https://cloud.csc.warwick.ac.uk/index.php/s/JomfpoMj87X3Jbk

Please let me know in case this does not work!

-Rudo

Rudo Roemer : 05 Dec 2019 16:29 |  Tags: Fastai Keras Machine-Learning Tensorflow |  Comments (4) |  Close comments |  Report a problem

Minimum requirements for publishing a numerical study using "blackbox" algorithms/software

Writing about web page https://www.journals.elsevier.com/physica-e-low-dimensional-systems-and-nanostructures

In my editorial work, I often encounter manuscripts in which authors are using more or less well-established numerical software packages to produce scientific results. These software packages may sometimes be of a commerical nature or already have a long development history. Typical examples of such codes are VASP, COMSOL, CASTEP, Siesta, GROMACS, etc.

Usually the authors of the manuscript are not developers of the software packages and simply use them as is. There is of course nothing wrong with this. However, in some case, these manuscripts contain (i) results only calculated for a single set of input settings, (ii) give numerical data without any indication of the accuracy of these estimates and/or (iii) any indication as to how these data are sensitive to the chosen parameter settings.

To be concrete, image a numerical DFT study of a certain material under strain and the determination of one of its lattice parameters as "a"=3.1234. Clearly, in many codes this number will depend on, e.g., the chosen ernergy cut-off as well as the number of k-points used in its basis set. In order to ascertain the accuracy of the chosen value for "a" one could, e.g., increase the number of k-points and observe how much "a" changes. Similarly, "a" usually changes when the cut-off energy changes, say from 500meV to 700meV. Both changes result in a new value for "a", say "a"= 3.1500 from the new energy cut-off and 3.1034 from the changes in chosen k-points. Indeed, one can usually get other values by chosing other k-point meshes and larger energy cut-offs. Hence the quoted final result should be something like

"a"= 3.1234 +/- 0.03 +/- 0.02 = 3.12 +/- 0.05

accompanied by an explanaton as to how these error bars were obtained. Only with these error estimates can readers of a scientific article see how accurate the data really is and how stable to variations in input parameters. Note that these values are still subject to further sources of error due to, (i) other input parameter dependencies and (ii) systematic errors that might be present in the chosen software package itself. Nevertheless, as given above, the bold final result at least provides some insight into the validity of the quoted numbers.

Rudo Roemer : 09 Jun 2019 18:33 |  Tags: #Physe Editorial Requirements Publishing |  Comments (0) |  Close comments |  Report a problem

Applying to review for Physica E

Writing about web page https://www.journals.elsevier.com/physica-e-low-dimensional-systems-and-nanostructures

As editor of Physica E, I often receive letters from younger colleagues asking to be considered as reviewers for the journal. My answer to this is usually as given below:

Dear ,

Thanks a lot for your interest in joining the Physica E pool of reviewers. Indeed, we are always keen to enlarge our group of experts to go out for a review. One of our key criteria is that these colleagues are known authors themselves in the field of specialization and have a publication record in the Physica E areas of excellence, i.e. nanostructures and low-dimensional systems. This excellence and engagement with Physica E’s research areas is usually best supported by previous publications in Physica E.

You can join our peer review pool by registering in EVISE to start the process, i.e.

https://www.evise.com/profile/#/non/login

In EVISE, as soon as a person registers in the system, they become available as reviewers; however, they will be able to access their reviewer profile/account of a journal only when they are invited by the journal for the review process.

Rudo Roemer : 03 Sep 2018 22:39 |  Tags: Physe |  Comments (1) |  Close comments |  Report a problem

A comment on: University strike puts final exams in danger

Writing about web page https://www.thetimes.co.uk/edition/news/university-strike-puts-final-exams-in-danger-l27b9x7vn

A comment on https://www.thetimes.co.uk/edition/news/university-strike-puts-final-exams-in-danger-l27b9x7vn

This is a fairly reasonably written article, thanks for that to "The Times". Some of the comments made by members of the public are rather less so.

VC pay is certainly a point that received attention recently, but it is not the main issue here. The UK university landscape is booming, but that has not resulted in marked increases in staff pay. On the contrary, the original final salary pension scheme, run by USS until 2016, that most academics joined at their universities, was downgraded to a less attractive defined benefits scheme (DBS) to make it "stable". Let me also point out that in distinction to most non-academic careers, lecturers will routinely spend years (~10) of their early employment in further study, fixed-term situations and fellowships, many of which do not result in pension-eligible contributions, before landing that first permanent employment (and many never do). A simple comparison with the "private sector" is hence missing the point. The reward for lecturers until recently was a "lowish" salary coupled with a good pension.

Some comments correctly ask why the previous scheme now seems in trouble. The question for academics is slightly different: why should they now trust a proposed defined contributions scheme (DCS), where you pay without knowing what you get, when the track record of the current USS scheme is so dismal? Furthermore, the current deficit is an actuarial one: for the purpose of the pension fund, all participating universities are assumed to be part of a single "company" and the "risk" to the pension fund is one in which all these places go bust at the same time. Times readers will know that this is a highly unlikely scenario (I do not know a case when this has ever happened).

Some commentators are correctly unhappy about the possible impact on student education and progression. As lecturer, I fully agree. As employee at a university, I am faced with effectively a substantial (future) pay cut by my employer. Hence it is my employer, the university, who has decided not to pay for the lectures (and the marking, the supervision, the tutorial/pastoral care, the work done in research, etc.). Let me also note that students have overwhelmingly joined forces with lecturers in this dispute.

Last, but not least, perhaps it is useful to point out that the universities involved in the current strike action are the older, more research active universities, i.e. many of the crown jewels of the UK university system. The younger universities of the post-90 age are part of the largely taxpayer-funded "teacher pension scheme" (TPS). Colleagues in that scheme tell me that their pensions are expected to be better than what USS will provide. The TPS scheme still contains elements of final salary and career-averaged benefits (since 2015).

Rudo Roemer : 20 Feb 2018 14:01 |  Tags: Ucu University Strike 2018 Uss |  Comments (0) |  Close comments |  Report a problem

Secretary of State for Exiting the European Union, response to my original letter

Follow-up to further response by D Dalton MEP from Rudo's blog

This is the reponse received via Mr Dalton's letter to the Secretary of State for Exiting the European Union (formatting is mine, but closely follows the original PDF):

3 May 2017

MR RUDOLF ROEMER, 17 THE RIDGEWAY, WARWICK, WARWICKSHIRE, CV34 5SH

Thank you for your letter of 24 April, regarding the UK's decision to leave the EU. I am responding on behalf of the Secretary of State.

Mr Roemer has raised a number of concerns, which I will answer in turn.

Residency/Citizenship

The Prime Minister underlined in her letter to the President of the European Council that we should seek an early agreement on the rights of UK nationals in the EU and EU nationals in the UK on a reciprocal basis. We have been clear that we should always put citizens first. This is a priority issue for the forthcoming negotiations.

We remain full members of the EU for the next two years, and subject to all the rights and obligations of membership. There will be no change to the rights and status of EU nationals living in the UK, or UK nationals living in the EU, while the UK remains in the EU.

The arrangements in relation to the movement of EU citizens into the UK after exit are, of course, matters that the Home Office is currently looking at, and they will be subject to discussion by Parliament.

We want to reach an agreement as early as possible. As soon as the other Member States have agreed the negotiating mandate for the Commission we want to start discussing this issue.
Please let me reassure your constituent that EU nationals who have lived continuously and lawfully in the UK for at least 5 years automatically have a permanent right to reside. EU nationals also have the right to reside in the UK with their family members and be treated equally to UK nationals if they are studying, working, self-employed or self-sufficient.

EU nationals who have lived continuously and lawfully in the UK for at least 6 years are eligible to apply for British citizenship if they would like to do so.

Work

The UK will remain an open and tolerant country, one that recognises the valuable contribution migrants make to our society and welcomes those with the skills and expertise to make our nation better still.

Social Security Benefits and Pension

The UK State Pension, along with other social security benefits is payable based on the UK National Insurance contributions that have been paid. Nationality or citizenship does not form part of the criteria for the award of State Pension.

Access to Education

The UK will always welcome genuine students and those with the skills and expertise to make our nation better still. We have already confirmed that existing EU students and those starting courses in 2016-17 and 2017-18 will continue to be eligible for student loans and home fee status for the duration of their course. We have also confirmed that research councils will continue to fund postgraduate students from the EU whose courses start in 2017-18.
The Government also recognises the important contribution made by students and academics from EU Member States to the UK's world class universities. A global UK must also be a country that looks to the future.
Please pass on my thanks to your constituent for taking the time to write and I hope that this answers Mr Roemer's questions.

RT HON DAVID JONES
MINISTER OF STATE FOR EXITING THE EUROPEAN UNION

Rudo Roemer : 09 May 2017 19:05 |  Tags: Brexit Eu Citizen |  Comments (0) |  Close comments |  Report a problem

further response by D Dalton MEP

Follow-up to Reponse by Daniel Dalton MEP (and Anthea McIntyre MEP) from Rudo's blog

Today I received a reply from Mr Dalton's contact to the UK government (formatting removed, "copy of the reponse" in next follow-up):

Dear Mr Roemer

Further to my last email of 12th April 2017, you recall I sent a copy of your email to the Secretary of State for Exiting the European Union.

As promised, I enclose for your information a copy of the response that I have received from the relevant Minister.

Yours sincerely,
Daniel Dalton MEP

Daniel Dalton
Conservative MEP for the West Midlands
Tel: 01926 930683
www.danieldaltonmep.co.uk
www.facebook.com/DanielDaltonMEP
Twitter - @ddalton40

Rudo Roemer : 09 May 2017 18:58 |  Tags: Brexit Eu Citizen |  Comments (0) |  Close comments |  Report a problem

Response by Kirty Murrel on behalf of Jill Seymour (MEP)

Follow-up to Brexit 2019, what I expect, letter to my local MP, Mr Chris White (Warwick and Leamington) from Rudo's blog

Rudo Roemer : 20 Apr 2017 11:27 |  Tags: Brexit Eu Citizen |  Comments (0) |  Close comments |  Report a problem