Hi all,
Information regarding the supplementary exam:
If the student has a supplementary exam that was approved through Special Consideration, they would have been informed by the Special Consideration team to check their exam timetable in myUNSW. This should be available 7 days before the exam, and includes the exam time and location.
—-
Please confirm the exam timetable for your supp exam, if relevant.
Regards,
Aditya
Dear all,
For those of you who participated in the industry project, here are the winners:
==
Technical Solution Winner: NLP_Team
NLP_Team demonstrated exceptional technical expertise with their comprehensive exploration of fine-tuning techniques. They implemented multiple innovative approaches including single-task and multi-task LoRA fine-tuning on modern models like Llama 3.2 3B and Gemma 3 4B, as well as partial layer fine-tuning and extending models with trainable heads. Their detailed experimentation and clear communication of results, accompanied by a working demo, showcased best practices in modern NLP implementation and evaluation.
Honorable Mentions:
Seek for $500, NetFineTuna, Coolest Team
Business Solution Winner: MA4
MA4 impressed with their thoughtful analysis of the business implications of various NLP solutions. Their submission featured an outstanding video presentation, clear articulation of the price-accuracy-speed tradeoffs, and meaningful fairness analysis. What particularly set them apart was their concrete recommendation for SEEK based on a thorough consideration of various business factors including control, customization, and data privacy needs—exactly the kind of actionable insights needed in real-world business contexts.
Honourable Mentions:
SeekDeeper, Data
(Note: the blurbs are AI-generated based on our evaluation notes.)
==
Congrats to all the teams mentioned above!
I will be in touch regarding the next steps. Please note that the requirement to receive the cash prize is to share your code with SEEK. All decisions regarding the winners above have been provided by SEEK.
Regards,
Aditya
Dear all,
If you have been approved for a supplementary exam, please read along:
- You are allowed to carry 1 A4 sheet handwritten on both sides. Calculators are neither required nor expected.
- The supplementary exam will be on Inspera as well.
- There will be a mix of multiple-choice questions and short-answer questions. There will be 8 multiple-choice questions (with sub-questions), and 4 short-answer questions. Negative marking applies to multiple-choice questions.
- Please read the instruction sheet carefully at the start of the exam.
The Exams team will get in touch with you regarding the exact date/time.
Best wishes!
Regards,
Aditya
P.S.: I expect to hear from the industry partner regarding project prizes by early next week. I will make an announcement in that regard.
P.S.2: All queries regarding marking have been addressed, as applicable.
P.S.3: We are not monitoring forum posts regularly any more. Please send an email to me with your queries.
Dear all,
Your group project marks are now available in Moodle. You will also see comments where applicable.
- If you submitted the group project but do not see your marks, please write me an email clearly mentioning your zID and team name .
- If you wish to seek clarifications about your marks, please write me an email Cc'ing all your team members. Please ensure you mention the zIDs of everyone along with team names.
I request you to raise concerns with respect to the above by Friday 3pm.
Regards,
Aditya
Dear all,
The final exam details are as follows:
|
Exam Opens: Thu 15-May 1:45PM
Exam Ends: Thu 15-May 4:00PM TOTAL TIME: 2 HOURS + 15 mins |
|
You should have received information regarding your exam venue from the UNSW Exams team. Please contact exams@unsw.edu.au soon if you have not. Note that the final exam is centrally managed and invigilated, and must be completed at the specified venue.
- The exam team has asked me to communicate with all of you that "the SEB for MAC is version 3.3.3. Click here to download
MAC 3.3.3
"
- The final exam has 15 multiple-choice questions and 2 questions with textual answers. Most questions have multiple subquestions.
- All questions/subquestions must be answered correctly to get full marks.
- Please read all instructions in Inspera carefully at the time of the final exam.
- The final exam is a hurdle to pass the course.
Good luck! :)
Regards,
Aditya
Dear all,
Hope the project submission went well. I would like to share a few things here:
(A) Final Exam
Please familiarise yourself with the Inspera Student Resource Site , where you will find videos, guides, resources and other information related to taking an exam on Inspera. If you have any queries about your final exam or final exam conditions, please contact the UNSW Exams Centre on exams@unsw.edu.au
Note that final exam will be managed by the UNSW Exams team. Therefore, it is best that you use the email above if you are facing issues (such as not being able to see the upcoming exam in Inspera, or not receiving the timetable for the final exam).
(B) Practice Exam
A practice exam has also been added to Moodle. Please note that we will not be marking the practice exam. It does not count towards the assessment for the course, and is, therefore, completely optional. However, I would strongly recommend that you try it to get an idea of the kind of questions to expect in the final exam.
(C) Assignment Marks are now on Moodle. You should be able to see them. The average score is approx. 21/25 - so you all have done really well as a class! :)
(D) myExperience survey is open until 1st May: https://go.blueja.io/DIp4NklFqUaTJH7kFAZAZQ . We have received 38% responses so far. If you have not done so, please complete the survey. Your comments will be greatly valuable to ensure that we keep running the course. As announced in week 9, if the response rate reaches 70%, I will be releasing additional information about the questions to expect in the final exam.
(E) Consultation sessions will not run hereafter. However, please email me to schedule time to address queries with respect to the final exam.
Thanks again for being a fantastic class, and good luck for the final leg of COMP6713!
Regards,
Aditya
Hi all,
Moodle issues seem to be resolved. You will be able to upload your projects to Moodle. The deadline is unchanged: Due: Monday, 28 April 2025, 8:00 PM .
Please do not send project files by email.
Late submission penalty as per university rules applies.
Regards,
Aditya
Hello all,
As a substitute for the lecture on Friday, the lecture recording for Week 10, Lecture 1 is here: https://echo360.net.au/lesson/933e76af-5362-430d-b693-011caf63a8c9/classroom
Please watch it before Thursday's class. Please bring your questions on the recording. I will also discuss the final exam. Additionally, we have an in-person guest lecture by Dr. Dai from CSIRO on Thursday.
I hope to see many of you in the last in-person lecture for COMP6713!
Regards,
Aditya
P.S.: myExperience survey is now live. I request all of you to please consider completing the survey. Your feedback is valuable.
Hello all,
This Thursday, we will have a talk by Dr. Xiang Dai from CSIRO. The speaker will deliver the talk in person. The talk will NOT be included in the syllabus for the final exam.
I will discuss details of the final exam and, as always, some lecture material - which is included in the syllabus.
Details of the talk are below:
----
Title: Active Adverse Event Surveillance: Are We Making Progress?
Abstract: Active adverse event surveillance aims to mine adverse drug events from a variety of data sources, including electronic health records, scientific publications, social media, etc. This is an active area of research, with many datasets being developed and shared tasks organized to support progress. However, one trend we’ve observed is that most of these resources and shared tasks focus on extracting adverse drug events from a single type of text. At the same time, there is growing excitement around large language models as general-purpose AI tools capable of solving a wide range of NLP tasks. This led us to take a step back and ask: How far are we from having a single adverse drug event extraction model that performs well across diverse types of text?
Bio: Dr. Xiang Dai is currently a research scientist at CSIRO Data61. His research interests lie in natural language processing and digital health. His work focuses on several areas, including extracting entities and their relations from unstructured text, pre-training domain-specific language representation models, and applying NLP techniques to the medical domain.
----
This lecture is the last opportunity for us to meet in the class. Hope to see most of you there!
Regards,
Aditya
[Typo in the previous version. Not 450 but 540]
Dear all,
Due to unforeseen circumstances, the tutorial in Rupert Myer 440 has been moved to room 540 at the same time. This is only for today.
Apologies for the inconvenience.
Regards,
Aditya
Hello all,
The lecture recording for Week 9, Lecture 1 is here: https://echo360.net.au/lesson/78e5de25-b478-4c62-845b-038a4e10747a/classroom
Please watch it before Thursday since the lecture will continue from this point. On Thursday, we will cover reinforcement learning for summarisation along with our guest talk as announced earlier. See you in class!
Regards,
Aditya
P.S.: myExperience survey is now live. I request all of you to please consider completing the survey. Your feedback is valuable.
Dear all,
During the next lecture (Thursday 17th April), we will have a talk by Dr. Ben Hutchinson from Google Research Sydney.
Note that the talk will be only a part of the lecture , and I will be covering substantial content myself. Also, as announced, I will be adding a recorded lecture to echo360 to make up for the public holiday. You are requested to watch the recording before Thursday's class.
Details below:
==
Title:
Towards Equitable Speech Technologies: Evaluating Speech Models for Biases
Abstract:
Evaluation of speech and language models is a critical part of the product development lifecycle, and better evaluation practices produce better models and better experiences for product users. It is increasingly accepted that robust NLP evaluation practices should consider the disparate performance of models for different groups, in order to produce fair and equitable outcomes for product users. However, in recent years many research projects have found that speech and language models perform worse for groups which are minoritised, and/or who speak minoritised language varieties. In the first part of this talk, I will discuss the evaluations of speech technologies which revealed discrepancies in model performance and real world outcomes for Black users of technologies in the US. In the second part of this talk, I will discuss a project that Google Research Australia has recently announced in collaboration with UWA, which aims to evaluate and improve the performance of speech technologies for Aboriginal and Torres Strait Islander people in Australia.
Bio:
Ben
Hutchinson
is a Research Scientist in Google's Research team in Sydney, Australia. His research includes learning from various disciplines to inform the responsible development of AI. Prior to joining Google Research, he spent ten years working on a variety of products such as Google Wave, Google Maps, Knowledge Graph, Google Search, Social Impact, and others. As well as his research, he works closely with product teams on responsible practices for developing machine learning data sets and models. He has a PhD in Natural Language Processing from the University of Edinburgh, and a BSc/BA in linguistics and mathematics from UNSW.
===
The contents of the talk are not in the syllabus for the course. The speaker has indicated that he will present the talk <u>in person</u> . - so I encourage you to attend in person. The talk itself may not be recorded as per the speaker's preference.
I hope that you are enjoying the course so far, and that you will utilise the opportunity to hear from Ben.
Regards,
Aditya
Hello all,
Unfortunately, one of the tutors is unwell today. If your tutorial is from 5-6pm in Rupert Myers building, room 540, you are requested to join the tutorial in room 440 at the same time.
Rupert Myers 540 5:00pm-6:00pm -------> Rupert Myers 440 5:00pm-6:00pm
This adjustment is only for this week.
Regards,
Aditya
Dear all,
Since both Fridays in weeks 9 and 10 are public holidays, there will be no tutorials on the day. Therefore, we have organised an additional tutorial on Thursdays of weeks 9 and 10 instead . The tutorial will be held in Mathews 214, from 5-6pm . I hope you will take advantage of the additional tutorial. As announced in week 1, the material covered in tutorials (as well as lectures/quizzes) is included in the syllabus for the final exam.
See you all in class tomorrow! :)
Regards,
Aditya
P.S.: Some of you have emailed me regarding an error in a question in the week for quiz 7. Please complete the quiz - I will be updating the reference answer and run an automatic re-grade in Moodle.
Hi all,
Message from SEEK (applicable to anyone in the class - not just those working on industry projects):
==
I noticed just yesterday Anthropic announced Claude for Education . Related to this, I think you and/or your students can create an Anthropic account and then request free credits for student project work here: https://www.anthropic.com/contact-sales/for-student-builders . That might be an easy way for students to include a proprietary model in their projects.
==
Regards,
Aditya
Dear all,
As you all know, the group project deadline has passed.
If you have registered your group, it would be useful to confirm that the information is correct in this sheet: https://webcms3.cse.unsw.edu.au/COMP6713/25T1/resources/110958 In case of discrepancies, please post a message in webcms or send me an email.
I have sent out emails to ALL students who have not registered for the group project with a random grouping for the project. Please ensure that you have a discussion with your team members, and send me the information mentioned in the email ASAP. We will be unable to provide any extension to the group project submission deadline. Therefore, doing it on time will get you started soon.
Regards,
Aditya
Dear all,
If you have not formed a project group and have not heard from me regarding a random project grouping yet, you are requested to email me by Thursday 5pm.
Regards,
Aditya
Dear all,
Group project registration is due by 7pm today. The form will close at the time.
If you opt for the industry project, email me the completed NDAs. Following verification, I will share the data files with you. Details in previous notices.
If you are yet to find a team, please reply to my post in the forum - that will enable you to reach out to people to pair up.
Regards,
Aditya
P.S.: Weekly quiz also closes at 5:25pm today. :)
Dear all,
Hope you have been enjoying the NLP lectures and tutorials so far!
Guest lecture: Dr. Terrence Szymanski, Head of Generative AI at SEEK, will be our guest speaker this Thursday. Terrence will talk about NLP in industry. The talk will not be recorded, and is not included in the syllabus for the course. Note that the talk is only for 30 minutes, and that the module on sentiment analysis will be our primary course content this week.
Group project: Thanks for those who have registered their groups. Reminder to everyone else to complete the group registration by the deadline. Some notes about group projects:
- If you need us to help you form a group, please complete this form: https://forms.office.com/r/Rwu9d5JeD1 You do not need to complete this form if you already have a group.
- You can register your project teams by completing the form here: https://forms.office.com/r/TsLHNSWPah
- If you opt for the industry project, please email the NDAs for all team members collectively as a single zip. Please name the zip as your team name. It is impractical for me to keep track of individual NDAs.
-On the topic of team names, team names are useful if they are unique. If you have registered your team, you may ignore this for now. But, if you are yet to register your team, it would help to apply a few minutes of creativity when coming up with a team name. :)
Individual assignment:
-The individual assignment is due on 21st March, and will need to be uploaded on moodle.
-Make sure you name the zip file as your roll number. Look up the submission requirements in the assignment description.
Quiz:
Please remember to complete your quizzes on time since we will not be making an exception for anyone unless you have an approved special consideration.
Regards.
Aditya
Dear all,
Following our discussion in the class, the group project scope and industry project documents are now available on webcms alongside the NDA.
The registration link remains the same:
Please use the following form to register your teams: https://forms.office.com/r/TsLHNSWPah
Regards,
Aditya
Dear all,
It's time to register your project teams for COMP6713! Please use the following form to register your teams: https://forms.office.com/r/TsLHNSWPah
You can pick ONE among the following options:
(a) You choose your project topic.
(b) You choose to work on the industry project. (All participating members will need to complete the NDA at: https://webcms3.cse.unsw.edu.au/COMP6713/25T1/resources/110590 )
Marking rubric and expected scope will be the same for both of the above options. Marking will be done solely by the course team at UNSW. SEEK will decide the winners for the industry project.
==
A detailed 'project scope' document will be shared at the expected commencement of the project. The document provides a structured framework for you to define the scope of your project. If you choose the industry project, you will still need to align your project scope as per the document.
Registration Deadline : 19th March, 2025, 7:00pm Sydney time.
Regards,
Aditya Joshi
Dear all,
Language models have indeed made several tasks easier than ever. You can now type “Write a poem about kangaroos in the style of William Wordsworth” in the ChatGPT tool, and it generates exactly that for you. How easy is that! I wonder what challenges a probabilistic poetry generator might face. Ah, implementing one will help me understand attention in Transformer much better!
The individual assignment for COMP6713 is now available to you in Moodle. Look for 'Individual Assignment' on the course page for the assignment description PDF and other files. The individual assignment carries a maximum of 25 marks. It contributes to 10% of the assessment in the course. The assignment must be completed on your own and without the use of Generative AI or external assistance.
Deadline: Friday, 21 March 2025, 9:00 PM Sydney Time
Assignment submissions will be in the form of a single .zip file named as your zID only. The submissions will need to be uploaded to the Moodle page by the deadline. I am happy to answer questions on the webcms forum. When posting them, please make sure you do not reveal details that may impact learning of other students.
Good luck, and hope you enjoy working on the assignment!
Regards,
Aditya
P.S.: Week 2 quiz is currently open in Moodle, and will run until next Wednesday.
Dear all,
Welcome to the 2025 Term 1 offering of COMP6713 : Natural Language Processing !
Over the next ten weeks, we will see how computer programs can process human language, which, by the way, did not begin with ChatGPT! The COMP6713 team consists of tu to rs (Lihua, Rashini, Duke, Zechen, Martin*), and me, Aditya: your lecturer and convener for the course. We will use WebCMS (https://webcms3.cse.unsw.edu.au/COMP6713/25T1 ) as the discussion forum, and Moodle for assessments.
Some quick reminders:
Our first lecture is on Thursday, 20th February 2025 from 14:00-16:00 in E19 Patricia O'Shane 104. See you there!
Regards,
Aditya Joshi (he/him)
aditya.joshi@unsw.edu.au