M2 INTERNSHIPS
2024–25

Coordination: Marc EVRARD, Thomas GERALD

Link to the procedure for creating and signing M2 internship agreements (in English).

Find an Internship

Introduction

You will have to find a public research laboratory inside or outside of Paris-Saclay or a private company doing research in our field of AI and data science. We will also share with you some proposals. Your internship has to last at least 5 months of full-time work. It must be financially compensated by law — see service-public.fr.

It takes place between February and August of your 2nd master year and must end before August 31. 

Your internship must go through an approval process. You need to contact us, and quite a few papers must be signed. Do not start before you get all signatures: you will not be insured and the time already spent cannot be taken into account. 

At the end of the internship, you will be asked to make an oral presentation and a report of a quality approaching that of communication at a workshop or a conference. The internship will be graded and is worth 30 ECTS.

Recommendations

The internships can include data analyses, method comparisons, and theoretical analyses (we discourage software development-only internships). 

The Graduate School of Computer Science's website holds an official internship listing here: https://gsweb1.dsi.universite-paris-saclay.fr/internship/list.

We encourage you to search on your own. Companies such as Microsoft, Google, Facebook, Thales, Safran, Orange, RTE, Datadog, Dassault, Schlumberger, BNP Paribas, Société Générale, Total, IFP, Renault, Guerlain, 4Paradigm, Airbus, Quantcube, Essilor, Unilever, NukkAI, Onera, GE Healthcare, Thomson, and research labs at Paris-Saclay (INRIA, CNRS, CEA) have been offering internships in the past to our students. One of the main places to look for internships is LinkedIn.
You can also subscribe to various mailing lists, such as Bull-IA (AI), Bull-i3 (general CS), ATALA (NLP), or Parole (Speech), where you’ll find internship ads. Also, get in touch with teachers or researchers whose fields you’re interested in.

Application

Please start searching now, and at least before November 15. 

Referent teacher (Enseignant référent)

You must choose (in addition to your tutor/internship supervisor) a referent teacher (a teacher from the master’s degree or your home school) who will follow your internship. See general provisions (in French). 

Contact the teacher you chose to obtain their agreement. Their role is to:

• Give you feedback on your project proposal

• Follow your progress

• Advise you in the event of a problem

• Help you prepare, attend, and grade your final oral presentation

• Grade your report

Internship Planning Suggestion

During your internship, you will:

• Carry out scientific work (not just engineering work)

• Propose original contributions (not just make incremental improvements)

• Include aspects relating to your M2 AI courses: e.g., artificial intelligence, machine learning, data science, language processing

Month 1

• Conduct a bibliographic review of the state of the art

• Identify an appropriate approach with original aspects specific to your contribution

Month 2

• Write a project proposal (ask for feedback from your internship supervisor and your referent teacher)

• Do preliminary experiments

• Obtain feedback from your colleague and tutor (possibly give an oral presentation)

Months 3 to 6

• Carry out systematic experiments comparing several approaches

• Analyze your results

• Gather feedback from your colleague and tutor (possibly give another oral presentation)

• Iterate over the previous steps

Month 6

• Complete the written report

• Prepare a final oral presentation to show your results to your work colleagues

• Internship Report (Aug. 31)

Format

Follow these recommendations:

• The intern must complete the form below before August 31 (23:59)

• Use the ICML latex template (template slightly adjusted for our Master) and follow the structure given below.

• 8 to 12 pages maximum of self-contained report in English + 1–2 extra pages of references + supplement material (e.g., extra tables, extra figures, code, systematic experiments)

Structure

• Headings

Title, name, GitHub repo URL + Challenge URL (as needed)

• Section 1: Background

Explain the context (answer the questions: what, why, what for). Review state of the art (answer the question: how). Motivate your contribution by identifying the limitations of current solutions and explaining how you contributed to tackling them.

• Section 2: Material and Methods

Describe the data you used (source, statistics about, e.g., number of samples, features, etc., visualize the data). This may include toy datasets, real-world application data, and standard benchmarks. Describe the general methodology employed, including how you monitored progress. Describe the ML techniques and/or data science models that you applied.

• Section 3: Results

Describe your results using enough tables and graphs, with confidence intervals (explain how they were computed), and compare them with baseline methods.

• Section 4: Conclusion

What the take-home message is, and what remains to be done.

Evaluations

Evaluation From Tutor (Sep. 3)

Your tutor (host institution) must fill the form in before August 31 (23:59) and send a copy of the Global assessment text field to: Marc Evrard and Thomas Gerald.

Here is the link you will have to forward to your tutor:

https://airtable.com/apppG049aAUtlXmUQ/shrpHzwWFjddIebrP

Report Grading

The tutor and other correctors may use the following guidelines to give a grade out of 20 points for the report.

1. Originality (5 points)
   Whether the proposed approach, analysis, and/or algorithms contain original ideas.

2. Scientific and technical quality (10 points)
   Whether:

   • The algorithms are implemented clearly and efficiently

   • Systematic experiments are conducted with comparisons with baseline methods and error bars

   • Good visualizations are presented

   • Results are critically assessed

3. Presentation (5 points)
   Whether the report is clearly written, well-presented with enough figures and graphs (referenced in the text and with good captions), and a bibliography.

Oral Presentation Grading

The tutor and jury members may use the following guidelines to give a grade out of 20 points for the oral presentation.

1. Problem Setting (4 points)
   Understanding and good presentation of the problem that arises, including a description of the data to be analyzed (if relevant).

2. State of the art (4 points)
   Literature review.

3. Method (4 points)
   The scientific and technical approach chosen, methodology: relevance and originality.

4. Results (4 points)
   Results and critical comparison with other approaches.

5. Presentation (4 points)
   Clarity of the presentation and charisma.