1 of 17

Projects

2 of 17

Project

Weight: 30%

All info available here!

3 of 17

Overview

All the dates have deadline at 23:59

Milestone P1

(individual)

Milestone P2

(team)

Milestone P3

(team)

10%

20%

70%

Due

Oct 3th

Due

Nov 7th

An outline of project ideas

Project proposal and initial analyses

Final project repository and datastory

Due

Dec 19nd

Three milestones: first glimpse, proposal, execution

02 Oct 2026

06 Nov 2026

Due

Nov 7th

18 Dec 2026

4 of 17

Datasets

https://go.epfl.ch/ada26_datasets

Provided datasets are available here:

5 of 17

Wikispeedia

Human navigation behavior data on a subset of Wikipedia, including:

Human navigation paths
Wikipedia hyperlink graph
Plain-text content of Wikipedia articles

Modalities: Text / Graph / Navigation behavior

Generic and versatile

6 of 17

YouNiverse*

Metadata about Youtube channels and videos

Collected for 14 years from 2005 – 2019, including:

Channel-level time series: weekly subscriber and view counts
Co-commenting graph

Modalities: Timeseries / Graph

Helpful dataset overview

*A very large dataset: please make sure that you have the resources to run whatever it is that you want to do on it

7 of 17

Reddit Hyperlink Network

Subreddit-to-subreddit hyperlink network extracted from posts that create hyperlinks from one subreddit to another

Network is directed, signed, temporal, and attributed
Each item includes:

Source and target subreddit
ID + timestamp of source subreddit
Post label: 1 if tone towards target is positive/neutral, -1 if it is negative
Post properties: vector representing the text properties of the post

Can be combined with embedding vectors dataset

Modalities: Graphs / Categorical / Temporal / Text

Helpful Dataset Overview

The hyperlink network represents the directed connections between two subreddits, extracted from publicly available Reddit data of 2.5 years from Jan 2014 to April 2017.

The subreddit-to-subreddit hyperlink network is extracted from the posts that create hyperlinks from one subreddit to another (i.e., there is a link in the source community that links to a post in the target community). Each hyperlink is annotated with three properties: the timestamp, the sentiment of the source community post towards the target community post, and the text property vector of the source post. The network is directed, signed, temporal, and attributed.

Each post has a title and a body. The hyperlink can be present in either the title of the post or in the body and therefore there is a network file for each.

It can be combined with a dataset of embedding vectors representing each subreddit

8 of 17

Stock Market Dataset

Historical daily prices for all tickers currently trading on NASDAQ (up to March 2020)

Temporal data for each of the NASDAQ traded tickers (5884 stocks and 2165 ETFs)
For each, values on: date, open, high, low, close, and volume

Modalities: Numeric data / Categorical / Time series

Helpful Dataset Overview

9 of 17

New Yorker Caption Contest

384 cartoon caption contests from New Yorker with ~1.4M captions and ~183M ranked votes

Aligned metadata includes:

Image locations, descriptions, wikipedia entities, caption ratings

Modalities: Text / Numeric data / Image

Can be combined with other sources (e.g. Wikipedia)

Helpful Dataset Overview

10 of 17

Multi-tissue COVID-19 Atlas

Collection of single or multi-omics datasets from 14 studies

Tissue is primarily PBMC (other tissues: kidney, nasal epithelium)
Focus: PBMC study

Contains 640k cells, 24k genes, 192 surface protein markers
Interesting metadata: Collection Day, Disease outcome, Age, Sex, Cell type, Smoker, Worst patient status

Modalities: Tabular numeric data (raw gene & protein counts)

Helpful Dataset Overview

Large single-cell atlas from COVID-19 patients and controls across multiple tissues such as blood, airway, and multiple organs. It contains a collection of single and multi-omics datasets from 14 studies.

However, we focus on one of these studies which is profiled with single-cell CITE-seq technique that simultaneously measures gene and surface protein counts in a single cell. There are 192 proteins with 24k genes and 640k cells.

This dataset enables cross-tissue comparisons of immune states across multiple levels of severity along with other metadata information such as ...

The provided AnnData file already includes PCA, Harmony, and UMAP embeddings for easier dataset visualization.

If you are interested in this dataset and do not have prior experience in working with single cell data, then it is recommended to take a look at how to process AnnData objects.

11 of 17

Visual Genome

Connects structured image concepts to language

Over 108K images, each annotated with:

Objects, attributes and pairwise relationships between objects
Region descriptions and question-answer pairs
On average: 35 objects, 26 attributes, 21 relationships per image

Modalities: Text / Image / Graph

Helpful dataset overview

Full Data

12 of 17

Cyclistic Bike Share Dataset

Bike-share trip data from June 27, 2013 to December 31, 2024

Each trip is anonymized and includes:

Trip start and end day and time
Trip start and end station
Rider type: Member (or Subscriber), Single Ride, Day Pass

Modalities: Text / Numeric data / Coordinates

Helpful dataset overview

Full data

13 of 17

Milestone 1

To be done individually, due 2nd Oct 23:59

No questions will be answered on Ed 24h before the deadlines!

1. Read the paper(s) relevant (Column G) to the dataset (fine if you don’t fully grasp the technical details about the method)

2. Play around with the dataset

3. Propose three bold and creative ideas (non-binding)

Deliverable: Outline of project ideas (up to 500 words)

Interesting examples from past years

14 of 17

Milestone 2

To be done as a team, due 13th Nov 23:59

A detailed project proposal (up to 1000 words)

A well-organised repository with initial analyses and data handling pipelines

15 of 17

Milestone 3

To be done as a team, due 18th Dec 23:59

The final repository: complete with all the analysis and findings

Provide a data story

Not a loose collection of results
Examples from 2017, 2018, 2019, 2020, 2021, 2022, 2023, 2024

16 of 17

Usage of LLMs

For Milestone 1, LLMs are not permitted.

For Milestones 2 and 3, please explicitly state in your report where LLMs were used.

17 of 17

Thanks for your attention!