Language is the most powerful tool created by human intelligence. We use language, our natural means of communication, to encode, store, transmit, share and manipulate information. In fact, most of the digital information available is unstructured information in the form of documents written or spoken in multiple languages, which represents a challenge for any organization that wants to exploit and process this information. Natural Language Processing (NLP)is thus at the center of our efforts to develop Artificial Intelligence (AI), and vice versa.



This subject is a follow-up of two subjects: Language Processing and Textual Data Mining. It focuses on the latest developments in NLP like, for instance, generative Large Language Models (LLM) like ChatGPT that are currently disrupting several areas of our societies. The student will be able to understand LLMs and use and evaluate them in specific applications. The course follows an active methodology, with group readings, presentations, discussions and a capstone project designed by the student.



The course is taught in English.

1. Understand the main concepts of large language models, and have a global vision of the area.

2. Understand and use large language models.

3. Understand the structure of applications that use large language models.

4. Analyze a specific application and identify the options and advantages of using large language models to deal with it.

5. Design and implement a specific application that uses large language models.

6. Design and carry out the necessary experiments to evaluate the performance of the specific application using empirical methods.



M011CE39 Ability and knowledge to design, build, use and evaluate robust systems for the NLP.

M011CE40 Master the different stages of development of an advanced artificial intelligence project: analysis, design, implementation, testing, presentation and evaluation of an advanced artificial intelligence project applied to a real problem.

This course introduces advanced PLN concepts and algorithms, focusing on large language models.



Topic 1: Language-centric AI. Introduction to empirical methods in NLP.



The first topic quickly reviews several topics, as follows: Scientific method applied to PLN. Knowledge-based systems, statistical systems, machine learning-based systems, and large language models. Supervised and unsupervised learning in PLN. Evaluation of PLN systems. Tasks and datasets. Benchmarks.



Topic 2: Advanced NLP techniques.



A review of Large language models. Creation, adaptation, prompts, alignment to user values and evaluation.



Topic 3: Advanced applications.



A review of the applications of large language models will be covered, including, among other, conditional text generation, question-answering systems, chatbots, summary generation, machine translation, etc.



In parallel to the theoretical topics, a practical capstone project is started, which will be developed throughout the entire course.

Through readings, presentations, assignments, and a capstone project, students will learn the skills necessary to design, implement, evaluate and understand a state-of-the-art NLP system.



The objective of the project is to be able to apply what has been learned during the degree to several applications, and to work the creativity of the students.



Active methodologies will be used throughout the course. That is, the students will participate actively both in class and in the laboratory.

• Continuous Assessment System
• Final Assessment System
• Tools and qualification percentages:
  • Realization of Practical Work (exercises, cases or problems) (%): 20
  • Individual works (%): 30
  • Team projects (problem solving, project design)) (%): 50

The UPV/EHU Protocol on academic ethics and prevention of dishonest or fraudulent practices in assessment tests and academic work will be applied.



The subject has two modes of evaluation: continuous evaluation and final evaluation.



The pre-registration in the continuous evaluation mode will be done at the beginning of the course. The pre-registration will become final after the confirmation of the application by the student on the dates established (between 60% and 80% of the course) and after verification of the partial performance by the teaching staff. If, on the aforementioned dates, the student does not confirm their final enrollment in continuous assessment, it will be understood that they renounce it.





CONTINUOUS ASSESSMENT



- 20% Realization of practices, cases or problems

- 30% Presentation and active participation in class: readings, bibliographic review, presentations and discussion.

- 50% Capstone project



All tests and activities mentioned for continuous assessment are mandatory.



The conditions to pass the continuous assessment are:

- Presentation and active participation in the seminar.

- Get 40% of the max. score in the practices

- Get 40% of the max. score in presentation and active participation in class.

- Get 40% of the max. score in the capstone project

- Get, after a weighted average, 50% or more of the max. score.



If the student does not express their waiver of continuous assessment within the given dates and does not meet any of the conditions to pass the continuous assessment, he/she will fail the subject in the ordinary call and may opt for the extraordinary call. If the student did not submit the final project, he/she would be assessed as a no-submit.







FINAL EVALUATION



20% Carrying out practices, cases or problems

30% written test

50% the capstone project



In order to take the written test, you must have previously handed in the individual practice. All the mentioned tests and activities are compulsory.



The conditions to pass the final evaluation are:

- Get at least 40% of the max. score in the written test

- Get 40% of the max. score in the practices

- Get 50% of the max. score in the capstone project

- Get, after a weighted average, 50% or more of the max. score.



If the student does not take the written test, it is understood that she waives the evaluation. If she/he did not submit the practices and project, he/she would be assessed as a no-submit.

FINAL EVALUATION



20% Carrying out practices, cases or problems

30% written test

50% the capstone project



In order to take the written test, you must have previously handed in the individual practice. All the mentioned tests and activities are compulsory.



The conditions to pass the final evaluation are:

- Get at least 40% of the max. score in the written test

- Get 40% of the max. score in the practices

- Get 50% of the max. score in the capstone project

- Get, after a weighted average, 50% or more of the max. score.



If the student does not take the written test, it is understood that she waives the evaluation. If she did not submit the practices and project, she would be assessed as a no-submit.

The material will be available on the virtual platform (eGela).

Basic bibliography

Jurafsky D., Martin J.H. Speech and Language Processing (3rd edition draft). An Introduction to Natural Language Processing, Computational Linguistics and Speech Recognition. Prentice Hall, 2018. https://web.stanford.edu/~jurafsky/slp3/



Lewis Tunstall, Leandro von Werra, Thomas Wolf. Natural Language Processing with Transformers, Revised Edition. O'Reilly Media, Inc. ISBN: 9781098136796

Web addresses

https://github.com/Hannibal046/Awesome-LLM