DOCUMENT

D r SamHall-McMaster is no stranger to new experiences. From a BSc (Hons) at the University of Otago to a PhD at Oxford University, he’s now doing further research at Harvard in the United States. He’s constantly applying his past experience to the new situations he comes across, and it’s this subject – how our brains enable the re-use of memory – that’s the focus of his current work. “This hasn’t been looked at in detail as it’s difficult to study due to the timeframes involved,” Sam says. “It’s a complex process, but I’mmotivated by the fact it’s something we do all the time.” This research into what’s known as generalisation is now amore realistic prospect due to improvement inmachine-learning analysis. “MRI scanning does an incredible job but it's still noisy data, whichmakes it challenging to draw conclusions,” says Sam. “Withmachine learning, we now have powerful algorithms that allow for more sensitive analysis.” This coincides with a shift in how brain activity is measured. Researchers used to focus on the changes in strength of response to a stimulus in a particular region of the brain. Sam says there’s now a focus on the pattern of brain activity as a whole. “This means we can deconstruct brain responses in a more sophisticated way.” Sam will assess this by scanning people’s brains as they choose between two images on a screen which might, for example, vary in colour or orientation. Images will be associated with a score ranging from 0 to 10 points. Over time, participants will get better at choosing the image that leads to a higher score, or ‘reward’. The critical test comes when a new set of images is shown – and this is where machine-learning comes in. “We can train a machine-learning model to learn the mapping between a particular pattern of brain activity and a particular piece of information,” says Sam. “This might be a decision rule like ‘choose red to get a reward,’ for example. Once we have a model that can distinguish brain activity based on those rules, we can see which information – which rules – are being reactivated when people are trying to make decisions about new images. In this way, we can assess whether people are reactivating the knowledge that previously led to good outcomes, or ‘reward’, when faced with a new decision.” The main region of the brain being analysed is the medial temporal lobe, which is essential for certain memory functions. It’s also known to be affected in neurological conditions such as Alzheimer’s disease. Sam hopes his work will give further insight into emerging research that suggests different subregions of the medial temporal lobe play different roles in using past experience. He hopes to ascertain whether experience is stored as a memory that is retrieved in its entirety, or if the experience is broken up into individual parts. He gives the analogy of a library. The lessons from specific past experience could be stored in a number of ‘books’ and catalogued for easy retrieval, so when a person is faced with a new scenario, they can draw on the exact knowledge required rather than having to consult all the books in the library. Sam gives the example of a person travelling to a different city and needing to use public transport. “Most people could easily do this, but it’s not a trivial problem,” he says. This is because the specifics will be different to those at home: the new city may have trains rather than buses, the station names won’t be the same and there may be a different ticketing system. However, it’s possible to draw on what we know about timetables, connections and payments. “We can solve the problem because we know transport is set up in a certain way. We recognise similarities,” he says. Isolating – and being able to retrieve – specific past lessons when faced with new problems provides a wide behavioural repertoire while still being efficient. In contrast, it would be wasteful if our brains had to retrievememories in their entirety every time. Sam says that a better understanding of how the medial temporal lobe’s subregions reorganise memories to influence flexible behaviour should help us understand neurological conditions like Alzheimer’s disease, in which it’s known that the medial temporal lobe undergoes structural and functional changes. This could allow for specific interventions as the disease progresses. This is a truly international project, Sam says. As well as working with Professor Samuel Gershman from Harvard University, he’ll be collaborating with Professor Nicolas Schuck from the University of Hamburg, and relying on funding from the Neurological Foundation’s Philip Wrightson Fellowship. He says the fellowship is critical to the research, allowing him to dedicate his time to this challenging issue. “This exciting global effort wouldn’t be possible without it.” Every day we retrieve stored memories and use them to navigate new situations – yet scientists know very little about how. Using the latest scanning techniques and machine-learning analysis, Neurological Foundation 2024 Philip Wrightson Fellow Dr Sam Hall-McMaster aims to find out. “With machine learning, we now have powerful algorithms that allow for more sensitive analysis.” SamHall-McMaster Headlines 15