HKPUG #99 - "Lie" to me, Python

Details

Date: 27 June 2026, Saturday
Time: 15:00 – 18:00 HKT
Venue: BC404, The Hong Kong Polytechnic University
Capacity: 40 participants

Please make sure you fill in the google form https://forms.gle/ogJSHfruCnjHNisU6 for Poly University QR Code Access.

Topics

1. Mind the Data Gap — Prize Ceremony & Winners’ Sharing

Speaker: Winning teams / participants
Duration: around 15 minutes

In our previous HKPUG workshop, participants worked in teams on a messy Hong Kong-inspired synthetic transport dataset.
The data was intentionally corrupted with missing values, outliers, inconsistent categories, URL-encoded station names, confusing delimiters, and hidden signals.

Teams had to clean the data and submit their work to a fixed black-box scorer to see whether their cleaning improved model performance.
In this session, we will present the prizes and invite the winning teams / participants to share their approaches.

We will cover:
- Prize ceremony for the workshop winners
- How participants approached the data cleaning challenge
- What kinds of hidden signals were found
- Common patterns and mistakes we observed
- What we learned from designing a gamified Python data workshop

No model tuning.
Just data rescuing.

2. Python "Lies": Things That Are Not What They Seem
Speaker: Alex Au
Duration: 15 minutes

Python often reads almost like English — until it does something that is technically correct but completely different from what you expected.

In this rapid-fire talk, Alex will uncover several Python-specific “lies”: late-bound lambdas in loops, mutable default arguments that remember previous calls, nested lists that secretly share the same rows, generators that disappear after one pass, the surprising meaning of for...else, and finally blocks that can rewrite a function’s ending.

Each example begins with a short “guess the output” challenge, followed by an explanation of the Python behaviour hiding underneath.

Things are not always what they seem.

3. LEPAUTE Framework — FSD-style Visual Perception with Lie Group Equivariant Attention
Speaker: Carson
Duration: around 60 minutes

LEPAUTE is a Python-based visual perception / machine learning framework originally motivated by Tesla FSD-style autonomous driving perception.
Autonomous driving perception is not just about detecting objects in a flat image. A model needs to reason about movement, rotation, position, scale, and viewpoint changes while still understanding the same scene consistently.

In this talk, Carson will introduce the motivation behind LEPAUTE, the core technical ideas, and how Lie group equivariant attention can be applied to 2D visual perception. The session will also include a live or prepared demo.
This talk is suitable for participants interested in Python, machine learning, computer vision, autonomous driving perception, mathematical structures in AI models, and research-oriented software projects.

Note: This is an independent technical sharing and does not imply affiliation with or endorsement by Tesla.

Rundown
15:00 – 15:30 Registration and networking
15:30 – 15:45 Opening and HKPUG updates
15:45 – 16:10 Mind the Data Gap prize ceremony and winners’ sharing
16:10 – 16:25 Break and networking
16:25 – 16:40 Python "Lies": Things That Are Not What They Seem
16:40 – 16:50 Break and setup
16:50 – 17:50 LEPAUTE Framework technical sharing and demo
17:50 – 18:00 Open discussion, networking, and closing

Who should join?
This meetup is suitable for:
- Python developers
- Data scientists and machine learning practitioners
- Students interested in AI, computer vision, and data science
- Developers curious about autonomous driving perception
- Open-source contributors
- Anyone interested in practical Python workflows
- Anyone interested in the Hong Kong Python community

Notes
- This is a community meetup and is free to attend.
- Seats are limited to 40 participants.
- Please RSVP only if you plan to attend.
- Campus access details will be shared with confirmed participants before the event.
- Please help us keep the venue clean and restore the room setup before leaving.

Special thanks to The Hong Kong Polytechnic University and Prof. Zackary Sin for supporting the venue arrangement.