The Computational Media Lab at the ANU focus on computational problems for understanding online media and their interactions with and among humans. We develop core methods in machine learning and optimization, and use them to formulate and solve problems in multimedia understanding and online behavior analysis.
A new machine learning system that styles your caption like master story-tellers do.
The craft of paper writing can be mastered using recipes.
I like cakes, and I enjoy reading logically lucid research articles. This post argues that research papers can benefit from explicitly thinking about and planning its four logical layers, just like a multi-layered Mille-feuille, or Napoleon cake (image credit: alyonascooking).Read More
How tricks from computer vision and deep learning can be used to accelerate planning algorithms
Planning algorithms try to find series of actions which enable an intelligent agent to achieve some goal. Such algorithms are used everywhere from manufacturing to robotics to power distribution. In our recent paper at AAAI ‘18, we showed how to use deep learning techniques from vision and natural language processing to teach a planning system to solve entire classes of sophisticated problems by training on a few simple examples.Read More
Siqi and Alex received sports and community accolades, respectively. The lab enjoyed two outings in town.
Sports is a prominent theme in Canberra. This post celebrates two of our competitive athletes, and looks back at two outings when the whole lab flexed our muscles and got a little bit wet.Read More
The Hawkes Intensity Process (HIP) inspired a small cascade of puns, and a research-team-in-uniform
read this before you email us
We look for new team members who value curiosity, critical thinking, and diversity in ideas, skills, and cultural perspectives.Read More
How much promotion is required, and why should one constantly promote?
“The fundamental scarcity in the modern world is the scarcity of attention.” – Herbert A. Simon.
Human attention is a limited resource and influencing the mechanisms governing its allocation is the holy grail of advertisement. Our ICWSM’17 paper applies the recent HIP popularity modeling to further examine popularity under promotion and answer questions such as:
supplying the missing link between popularity and promotions
One major gap in understanding social media is to precisely quantify the relationship between the popularity of an online item and the external promotions it receives. Our recent WWW’17 paper supplies the missing link. We use a mathematical model to describe the continuous interaction between external promotions (e.g. tweets about a video) and popularity dynamics (e.g. daily views). This in turn answers several practical questions:
We analyze news sources on YouTube to reveal their roles in broadcasting information.
We are in an era when information consumption and production are democratized. Regular users not only consume information, but they digest, mutate and produce new information which gets passed on to other users. Prompted by a press inquiry from the Polish online news www.press.pl, we analyze the impact of these emergent sources of information versus traditional media, in the context of politics. More precisely, we study how YouTube videos posted by two traditional news sources (BBC news/UK, ABC news/USA) and two emergent online news sources (The Alex Jones Channel, The Young Turks) are viewed around the time of the US political elections of 2016.Read More
This post outlines techniques for computing the expected event rate for Hawkes processes, or the so-called Hawkes Intensity Process (HIP).
This post is the first of a series on modeling social media popularity using the Hawkes Intensity Process. “Expecting to be HIP (II)” gives an overview of results and interpretations on a large YouTube video dataset. “Expecting to be HIP (III)” further quantifies the effects and interpretations of promotion.Read More
We discuss three approaches for graph inversion from steady state distribution of random walks. This first post compares the problem settings.
Markov chains and random walk on graphs are long-standing subjects in applied math and related disciplines. Given a markov chain / transition matrix / graph, properties of the random walk such as steady state distribution, mixing time are well-studied. Only in recent years, however, the inverse problem has started gaining interest – that is, given some measurements after random walk on a graph, infer some version of the graph. This blog post compares three related but different problem settings in recent literature from
We explore Australian caves and cherries in the second half of 2016.
Australia has great weather for the ourdoors, no matter it’s winter or summer – that’s what we did, and here’s photo proof.Read More