The deadline is April 23rd at 10:00 Saarbrücken standard-time. You are free to hand in earlier. You will have to choose one topic from the list below, read the articles, and hand in a report that answers the assignment questions but above all, critically discusses the material beyond these. Reports should summarise the key aspects, but more importantly, should include original and critical thought that show you have acquired a meta level understanding of the topic – plain summaries will not suffice. All sources you use should be appropriately referenced, any text you quote should be clearly identified as such. The expected length of a report is between 3 to 5 pages, but there is no limit.
For the topic of your assignment, choose one of the following:
Reshef et al. [1] introduced a novel measure of correlation (or dependence) to detect associations in large data sets. They present their findings very confidently as the next big thing. Among others, Simon & Tibshirani [2] and Kinney & Atwal [3] think differently, and wrote rather strong rebuttals. Reshef et al. [4] did not quite agree with these rebuttals, leading to yet another answer by Kinney & Atwal [5], and ultimately to the work by Reshef et al. [6].
Who is right? Are Reshef et al. [1] presenting false claims and over-selling the results? Are Kinney & Atwal [3] purposefully mis-interpreting Reshef et al. [1] and ignoring their claims? Are Simon & Tibshirani [2] presenting sensible criticism, or are they pedantic and besides the point? What is your opinion, is the concept of equitability a useful one, and is MIC a useful measure? To what extent are MIC and 'equitability' useful concepts for when we want to explore (mine) and learn from data?
What does this exchange of letters and notes tell about the process of doing science? Has the general public been mis-led by the tone and impact of these publications? Was the Science magazine wrong at publishing the work of Reshef et al. [1] so soon? Was it scientifically acceptable for PNAS to publish the comments of Kinney & Atwal [3], or might attention and money be part of the equation for either publisher? What is the value of pre-print servers such as arXiv (where Simon & Tibshirani [2] and Reshef et al. [6] are published)?
Your report should answer to both the technical questions and the above questions about the process.
To do well on this assignment, you must have a good understanding of deep learning. Do not choose it because omg so hype!
The most talked-about machine learning topic of the decade is definitely deep learning. Many claim impressive results, such as classifying images [7], plaz GO [8], achieve scientific breakthroughts [9], generate wonderful images from text [10], and are nowadays used for everything but the dishes.
Take a moment, and think about a deep neural network truly is. What makes it deep? How do these applications really use this 'deepness'? Do they all use the same "deep learning"? Are each of these papers truly (only) about deep learning? For example, what about AlphaGo? What about you credit the deep net for the discovery of the new antibiotic, or, is there more to it?
Are all of these resounding success stories? Read Khurshudov [11] and Marcus [12]. (Udandarao et al. [13] is optional.) What is this wall Gary Marcus talks about? Dall-E 2 [10] was announced mere days after Marcus [12] wrote about the wall, and meanwhile Claude, ChatGPT 5, Gemini and DeepSeek are doing amazing things on a daily basis. Did we break through the 'wall' before it was erected, is the wall still there, as some argue, is there no wall to begin with?
Finally, how and to what extent does the predictive nature of deep learning clash with the descriptive goal of knowledge discovery? How can we gain insight into data? How can we gain insight into models? Is deep learning the answer? Or, are we bound to other tools?
Read the following three papers by Shahaf & Guestrin [14], Shahaf et al. [15], and Hope et al. [16]. Each of these considers a very different but also fascinating topic. Besides sharing an author, they also share a key aspect. What are the similarities, differences, and non-trivial connections between these papers? The answer is obvious once you know it. Think deeply!
To help you get started, consider the following example questions. To what extent do these papers, the methods, and their results convince you? Why? Are the goals equally clearly defined? Why? Are the choices principled or rather ad-hoc? Why? Is the evaluation convincing? Why? Are the results convincing? Why? Are there any key experiments missing you think would have been doable and necessary? Why? Can you identify possible improvements for the algorithms, how would you approach these problems?
Return the assignment by email to vreeken (at) cispa.de by 23 April, 1000 hours.
The subject of the email must start with [TADA].
The assignment report must be returned as a PDF with the '
You will need a username and password to access the papers. The first lecture gives you the password.