scholarships at A2I2
offers a number of full PhD scholarships (3-3.5 years). The
scholarship generally covers tuition fees and living expenses.
The Applied Artificial Intelligence
Institute (A2I2), co-led by
Venkatesh, is an Human-centred AI research body,
currently undertaking research in the following areas:
We study all relevant aspects of intelligence and
build systems that learn to read from text, perceive the world, build
remember the past, imagine the future, reason, collaborate with others
and act on the world. See this
blogpost for our view on machine learning, and its
recent developments. Major sub-areas:
» Grand unified cognitive architecture
in what we aim to build a single principled computational architecture
that can replicate and explain human cognitive capabilities across
scales in time. The ultimate goal is exploring the limits of computational approach to self-awareness and consciousness.
learning 2.0 in
that we find efficient ways for differentiable
with minimal human supervision, towards System 2 capability. Our
sub-areas include, but not
limited to: generative models, graph neural networks, relational
models, memory and
attention, continual learning, machine theory of mind, and learning to
reason. See this
blogpost on why deep learning works, this
post for a tutorial, and this page for
works done so far by us.
» Reinforcement learning
in that an agent acts on the world, interacts with others, builds
theory of mind, imagines the future and receives feedbacks. Equipped
with deep nets for perception, memory, statistical relational learning,
and reasoning capabilities, we aim to
bring RL to a new level.
» Build the world model
Learning from large labelled data as we do today does not scale, and it
is not anywhere near human intelligence. A better way is learning
to represent the world just by perceiving the world, reading from text
or watching videos with little hints from teachers. Intuitive
physics, knowledge graphs and the like will naturally emerge, just like
those discovered by infants.
» Value alignment
in that we build intelligence and ensure that the machine respects
human preferences, even if when the machine is capable of devising its
own goals and aware of what it does. The day when machine and human
co-exist and co-create seems far, but it is already here in every
fabric of life.
want to use AI to make the world better. Here are a few things we are
» Computer vision
which we want to understand and reason about videos, talk about movies,
read children stories, monitor
streets, and understand human behaviors in video. See iCetana for our product in action to
make a safer world.
» Natural language
in which we aim to build agents that read text, answer multi-turn questions,
hold a long conversation, share human values and talk about health. We
also aim to leverage NLP techniques for scientific extraction from text and ideas
in which we learn to read the scientific texts, extract machine
computable knowledge, model the emergence and dynamics of ideas over
time, and support scientific reasoning and question answering.
» Healthcare in which we aim to
(a) understand about patient's health, need,
preference and intention
through patient's diverse sources of data (medical records, -omics,
user-generated media, medical imaging and biosignals);
(b) acquire and reason about established medical knowledge;
(c) have a
meaningful dialog with patients; (d) recommend the personalized course
of medical actions; (e) support doctors and hospital managers to
improve their precision and efficiency. See this
blogpost for some futuristic ideas and this page for
works done by us so far. See our product Toby PlayPad for supporting
learning in autistic children.
» Drug design
in which we aim to model the entire drug-like space, compute
bioactivity, predict binding to proteins, and best of all, generate new
drugs with a given set of desirable properties.
» Structural bioinformatics in
which we aim to unlock the one of the most mysteries of nature -
our genomes, map genotype-phenotype, answer any genomic queries for a
given sequence (DNA, MtDNA, RNA, etc), predict drug-protein
interactions, and learn to generate proteins and DNAs.
» Accelerating scientific
which we use AI to replace hard physical computation and
experiments such as computing quantum chemical properties, predicting
molecular properties & interactions, predicting chemical reactions,
understanding the structure and characteristics of materials, searching
for new alloys, and generating molecules & crystals. Part of these
is accelerating new
by finding best configurations for producing new products. See our
recent work on a faster way to make new materials, and other works here.
» Smart homes
in which we model human activities and sensing systems/IoT within a
to assist/empower the tenants in their everyday life. An important AI
goal is to build a situated conversational agent that can hold
meaningful conversations with tenants. A high impact environment will
be aged care homes.
» Automated software
in what we want to read the code, fix the bugs, synthesize programs,
the programming process, understand developer and support team
management. See this
page for the latest update on our research.
cool apps: (A) Cybersecurity
detection in that we detect unusual behaviors
without supervision. We aim to handle the complexity of data types,
networks, relations and dynamics. (B) Recommender systems
in that we deliver personalized content to user. See this
post series for some of our works.
is no single selection criterion and we can only say that we look for
candidates who have good potential to have a
successful research career,
not just a PhD degree.
If you do not know what it takes for a research career, I recommend
that you should spend time to read about it. See, for example, this
blogpost for a view on how AI research these days is going.
Sometimes it will help you
decide whether you need a PhD or not. Here is the list of things
we are looking for
of knowledge and willing
to pursue a research career. PhD is just a stepping stone, and a
by-product of the learning process.
experience (e.g. has published papers).
recognition: university prizes and awards. High quality honours at high
school (such as members of national or international
teams) may also count.
in CS/Maths/Physics with excellent
academic performance from a well-recognised university. For those with
Australian qualifications, the standard requirement is an Honours
math background (we will use a lot of statistics, optimization and
programming skills. An ability to scale up to large datasets is a bonus.
machine learning & AI
background. As a general rule of thumb, a good AI/ML PhD student
will be likely to have covered statistical
machine learning & probabilistic
graphical models. A prior exposure to deep learning is a
English (generally IELTS score 6.5, no bands below 6, is
required for entry, but
you should be aware that this score is insufficient for smooth
communication). Good critical reading skills are vital
because you will spend up to 50% of your time for reading. Technical
writing is also important because it is often the end-product for
assessment, and it accounts for 50% chance of acceptance.
What to send
rule of thumb is that you should send whatever (electronic) documents
you think that will support your future research career. Before you
send these, I recommend that you spend time to study research areas and
publications by A2I2. Documents are written in
English and may include
some of the following:
comprehensive resume (or CV);
papers or technical reports (if written in languages other than
English, titles and abstracts should be translated to English);
statement of purpose briefly describes your research intention and
interests, general approach to reach your PhD/career goals;
» Proposal for the intended thesis; and
» Reference letters from your professors.
can send your documents directly to A/Professor Truyen Tran.