Due to the evolving situation in Boston, the format for the student research talks has changed. We will no longer be using the Youtube live-stream of the auditorium. Presenters will deliver their talks virtually via Webex screen-share/video conference.Here's the link to the Webex meeting. The password is bcstalks2020. 

Presenter: Andrew Francl, McDermott Lab

Title: Deep Neural Networks Reveal Adaptedness of Human Sound Localization to Real-World Environments

Abstract: Perception is believed to be adapted to the world. Yet adaptedness is often difficult to test. We propose to use contemporary machine learning to investigate adaptedness by revealing the properties of systems optimized for particular tasks. We explored this approach for human sound localization. We equipped a deep neural network with simulated human ears and trained it to localize sounds in a virtual environment. The resulting network exhibited many documented features of human spatial hearing when tested in simulated experiments. We then leveraged the ability to perform experiments on the model that are impossible with biological organisms, training it in unnatural conditions to simulate development in alternative worlds. These experiments suggest that the characteristics of human hearing are indeed adapted to the constraints of real-world localization, and that the rich panoply of sound localization phenomena can be explained simply as consequences of this adaptation.

Presenter: Gwyneth Welch, Tsai Lab

Title: Exploring early pathological signatures of neurodegeneration.

Abstract: The accumulation of DNA double strand breaks (DSBs) in neurons is associated with aging, cognitive decline, and neurodegenerative phenotypes. However, the mechanisms that link DNA DSBs to the development of neurodegenerative disease remain poorly understood. In a mouse model of severe neurodegeneration, we utilize fluorescence-activated nuclei sorting followed by RNA sequencing and H3K27ac ChIP sequencing to define transcripts and gene regulatory elements altered in neurons with pathological levels of DNA DSBs. Interestingly, these data reveal a subset of neurons with extreme genomic instability activate innate immune signaling pathways. Furthermore, utilizing a single-nucleus RNAseq dataset from Alzheimer’s Disease (AD) and non-AD individuals, we find this transcriptional signature is conserved in a subset of human neuronal nuclei. Finally, we find that inhibiting innate immune transcription in neurons can mitigate microglial proliferation and inflammatory gene expression, suggesting neurons with pathological levels of DNA DSBs contribute to neuroinflammatory phenotypes associated with neurodegeneration.

Presenter: Kelsey Allen, Tenenbaum Lab

Title: Rapid trial-and-error learning with simulation supports creative physical reasoning and tool use

Abstract: Many animals, and an increasing number of artificial agents, display sophisticated capabilities to perceive and manipulate objects. But human beings remain distinctive in their capacity for flexible, creative tool use -- using objects in new ways to act on the world, achieve a goal, or solve a problem. To study this type of general physical problem solving, we introduce the Virtual Tools game -- an online game where people solve a large range of challenging physical puzzles in just a handful of attempts. We propose that the flexibility of human physical problem solving rests on an ability to imagine the effects of hypothesized actions, while the efficiency of human search arises from rich action priors which are updated via observations of the world. We instantiate these components in the ``Sample, Simulate, Update'' (SSUP) model and show that it can solve problems creatively in similar ways to people. More broadly, this model provides a mechanism for explaining how people condense general physical knowledge into actionable, task-specific plans to achieve flexible and efficient physical problem-solving.

Presenter: Martin Schrimpf, DiCarlo Lab

Title: Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs

Abstract: Deep convolutional artificial neural networks (ANNs) are the leading class of candidate models of the mechanisms of visual processing in the primate ventral stream. While initially inspired by brain anatomy, over the past years, these ANNs have evolved from a simple eight-layer architecture in AlexNet to extremely deep and branching architectures, demonstrating increasingly better object categorization performance, yet bringing into question how brain-like they still are. In particular, typical deep models from the machine learning community are often hard to map onto the brain’s anatomy due to their vast number of layers and missing biologically-important connections, such as recurrence.

Here we demonstrate that better anatomical alignment to the brain and high performance on machine learning as well as neuroscience measures do not have to be in contradiction. We developed CORnet-S, a shallow ANN with four anatomically mapped areas and recurrent connectivity, guided by Brain-Score, a new large-scale composite of neural and behavioral benchmarks for quantifying the functional fidelity of models of the primate ventral visual stream. Despite being significantly shallower than most models, CORnet-S is the top model on Brain-Score and outperforms similarly compact models on ImageNet. Moreover, our extensive analyses of CORnet-S circuitry variants reveal that recurrence is the main predictive factor of both Brain-Score and ImageNet top-1 performance. Finally, we report that the temporal evolution of the CORnet-S "IT" neural population resembles the actual monkey IT population dynamics.

Taken together, these results establish CORnet-S, a compact, recurrent ANN, as the current best model of the primate ventral visual stream.

Presenter: Morteza Sarafyazd, Jazayeri Lab

Title: Computational and neural principles of inferential reasoning in the brain

Abstract: My research addresses the computational and neural principles of inferential reasoning in the brain. In our daily life, we often have to find the cause of our errors to make better decisions. In simple decision-making tasks, it is easy to infer the cause of errors. However, when the decision involves multiple if-then stages, errors can have multiple causes, and we have to rely on more sophisticated reasoning strategies to infer the cause. We developed a decision-making task for non-human primates to investigate how the brain makes such sophisticated inferences. Our behavioral experiments revealed that monkeys, like humans, integrate information over multiple timescales to disambiguate the cause of their errors. Next, we performed electrophysiology experiments to understand the underlying neural mechanisms. We found two areas in the frontal cortex that carried information about the animals’ belief about the various sources of error. Further causal experiments revealed how interactions between these areas enabled the animals to integrate information over multiple timescales and make causal inferences.