Welcome, the chances are that you are here, reading this page, because I’ve redirected you to this useful repo. It’s here to help and speed up your journey to a research career, where if appropriate could well be in collaboration with me.
This is a collection of useful hints and tips for postgraduate research, such as doctoral or masters study. Over the last 20+ years I’ve been fortunate enough to guide amazing people through their research journey, so here are my thoughts to help people in the early stages of their journey. This includes people considering collaborative research with me, so lists areas of potential mutual interest.
There is one other criteria I look for when committing to guide/supervise an emerging researcher for over three years: ‘Do they make me think?’ I’m interested in someone with an enquiring mind, who naturally asks the difficult questions.
There have been a few things that have made the journey more difficult, such as lack of finances, part-time study, starting a family, illness, but none of these have been impregnable barriers to completion. I can only recall two actual barriers: money as the motivator (entrepreneurship is to be celebrated, but rarely a good basis for research) and drugs (even soft drugs impair intellectual ability).
Academics receive hundreds of enquiries per year, so please do not be disheartened if an enquiry goes unanswered as it’s not personal. Thus, The first hint is to make the email personal, i.e. Dear Prof Will Browne rather than Dear Professor, (BTW the title can be left off for me). The second hint is to show that you have looked into the research of the supervisor, e.g. I read your paper on … [and have actually read it!]. The third hint is to show how your research interests overlap with the supervisor’s current interests, e.g. This matches my interest in …, which I developed from … Finally, it is good to show the direction of the joint research, e.g. Would you be interested in guiding my research on … ?
Commenting on the latest research trends, areas of future investigation, pertinent research questions and hypotheses to address open questions, will also get the interest of potential supervisors.
CVs should be quantifiable, e.g. put your grades in relevant subjects, overall GPA and English-language test scores. If your university is not in the same country, then it’s good to cite its world ranking and how it compares nationally. Published papers are often important, so state the impact factor of the venue and its number of citations. Academic CVs do not have to be restricted to one-page, but they do have to be nicely formatted and free of typographical/grammar/spelling errors.
This is not an exhaustive list, but is indicative of my research plans. It’s also not in a priority order as it’s more important to find areas of mutual interest rather than adapt to a supervisor’s area of interest that is not motivating to you.
My overall research area is: Artificial Cognitive Systems
That is, how to get artificial systems (software and/or robots) to perceive, represent, reason, learn and effect an interaction with an environment to continually improve its task solving ability.
AI that is pre-programmed is limited on its tasks and human bias. Learning systems offer richer behaviour, where collaborators and my team has created the following that need developing (projects bulleted)
A symbolic system that works in Boolean and Integer domains without noise
A lateralized system exists that enables an input to be considered at the local level and the holistic level simultaneously.
A compaction algorithm exists that can remove redundant and irrelevant learnt knowledge, where it is unknown how this works for continual learning. What happens with almost duplicate knowledge from different sources, how to forget (without catastrophic forgetting), how to learn economically and so forth.
This is the ability to learn the underlying patterns within an environment to interactive observation within it. This goes beyond generalisation, i.e. the ability to identify and ignore irrelevant/redundant knowledge. Instead, it is the ability to identify building blocks of knowledge that can be recombined into more complex patterns that are useful to solve tasks in the domain.
The continual learning system developed is a basic form of abstract reasoning in terms of recombining solutions from small problems to help solve larger-scale problems in the same or related domain.
Robots are often required to produce maps of unknown areas or navigate through dynamic environments but use a static map of fixed scale. Humans and other animals appear to be able to switch scales in a map to reason at the best scale for the problem they are tackling currently, e.g. local scale to move around an obstacle or more globally the best path to achieve multiple objectives.
Story telling appears fundamental to what it is to be human, especially as a basis for memory and intelligence. It is fascinating that we can not remember our early years of existence, which are arguably our most formative. For many cultures memory starts between 6-8 years old, yet in Maori culture it is between 2-3 years old. It has been postulated that it is the aural - story telling - tradition (combined with the weight placed on significant family events such as weddings and funerals) that contributes to the ability to recall impactful events. Dave Goldberg sums up story telling exquisitely “ … stories as the central way in which human beings approach complexity. Stories contain, time and spatial relationships, causality relationships, intentionality relationships, clues regarding ontological modes & epistemic certainties and uncertainties, all in a compact representation with the possibility of pointers to other stories as well as visual and mathematical representations.” [https://www.linkedin.com/pulse/3-lessons-i-learned-from-john-h-holland-david-e-goldberg/?trackingId=gtpv22IKSaGXIQ2gKkCGdA%3D%3D]
This project seeks to create a robot that can experience its world and tell stories about it.
As an aside, Rodney Brooks is brave enough to make predictions about the future and revisit them yearly, where he poses: “An AI system with an ongoing existence (no day is the repeat of another day as it currently is for all AI systems) at the level of a mouse.”, where a storytelling robot could be a step in this direction, albeit it would need to be a temporal step in a journey that may or may not have a happy ending.
It encompasses many hard problems: “symbolic grounding”, “the Frame problem”, “intentionality”, “narrative” and many others. But with computer vision for segmantation, semantic maps, large language models, rule-based symbolic learning, embodied robotics, and large compute power, we may be able to get a robot to coherently answer ‘so, what did you do on your holidays?’.
Reasoning using symbols is powerful when describing innate patterns of features in a domain. It can also form the basis of eXplainable AI (XAI), which will become an essential topic for adoption of AI in every day decision-making. However, presenting such systems with raw data, e.g. pixels in an image, phonemes in sound, characters in a sentence and so forth overwhelms symbolic reasoners as there are simply too many potential patterns to consider for the computational resources available. Instead, connectionist approaches, especially deep neural networks, have been shown to be excellent at determining important features and their relationships to environmental targets. Such end-to-end reasoning is impressive but has also been described as a locality sensitive hashing table. It doesn’t understand the underlying patterns.
Marvin Minsky, a pioneer in AI, raises the question whether a machine can be intelligent without emotions rather than having an intelligent machine expressing emotions. Affective robotics considers how analogues of emotion can be used to improve performance, especially in uncertain, unknown, dynamic and noisy domains where look up-table type approaches are likely to fail to discover identical previous experience. It is noted that this field is controversial as the neuroscience basis for emotion itself still requires much research. Previous research has developed ‘hidden latent variables’ that that can be adaptively learnt to enable functionality equivalent to that considered useful in emotions.
Much of my research is forward-looking, blue-sky ideas, so it is worthwhile to consider a beneficial application. Namely, intelligent orthotics for stroke rehabilitation that learn the best strategies to help people affected by stroke. A prototype hardware system has been constructed, which ironically will take time to develop for medical approval.