Please join us for a series of research talks from students in the BCS graduate program. 

Date/Time: Friday, March 10th from 4pm - 6pm

Location: On Zoom https://mit.zoom.us/j/95673653804

Presenters:

Mahdi Ramadan, Year 6 Graduate Student in Mehrdad Jazayeri’s Lab

Title: Macaques can reason counterfactually

Abstract: A hallmark of human intelligence is the ability to consider counterfactual explanations for past experiences. It is not known if this capacity is uniquely human or present in other species. Here, we address this question by analyzing the behavior of rhesus macaques in an intuitive hierarchical decision-making task. Monkeys saw a ball enter a maze and disappear. While invisible, the ball moved inside the maze toward one of several possible exits. Monkeys received temporal information every time the ball changed direction and had to use this temporal information to infer the correct exit. Comparing behavioral responses to a collection of cognitive models implementing different inference algorithms revealed that monkeys improved their decisions by evaluating counterfactuals. This finding was robust to experimental variations including novel maze geometries and perturbation experiments. High-density neurophysiology recordings from two brain areas support these findings. These results trace the roots of counterfactual reasoning to distant points in the primate lineage.

Mitchell Murdock, Year 6 Graduate Student in Li-Huei Tsai’s Lab

Title: Frequency-specific vasoactive neuropeptide release regulates glymphatic clearance 

Abstract: The glymphatic movement of fluid through the brain powerfully clears metabolic waste. How neural rhythms regulate glymphatic transport is incompletely resolved. We observed sensory gamma stimulation increases the influx of cerebrospinal fluid and the efflux of interstitial fluid in the cortex of a mouse model of Alzheimer’s disease, which was associated with increased aquaporin-4 polarization along astrocytic endfeet, dilated meningeal lymphatic vessels, and amyloid accumulation in cervical lymph nodes. Inhibiting glymphatic clearance via genetic, pharmacological, and anatomical interventions abolished the removal of amyloid by gamma stimulation. Using chemogenetic manipulation and a novel genetically encoded sensor for vasoactive intestinal peptide (VIP), we found VIP+ interneurons facilitate glymphatic clearance during gamma stimulation by regulating arterial pulsations. Our findings establish novel mechanisms to recruit the glymphatic system to remove brain amyloid.

Setayesh Radakani, Year 4 Graduate Student in Rebecca Saxe’s Lab

Title: What people learn from punishment

Abstract: When a parent or a judge chooses to punish, they often intend to show that, and how much, the punished act was wrong. However, in light of every act of punishment, targets and observers evaluate not only the action that elicited the punishment, but also the motives and legitimacy of the authority who punished. Both in real life and laboratory settings, the same punishment can lead to contrasting and even contradictory consequences in terms of changing others’ beliefs about undesirability of the act, as well as the motivations and legitimacy of the authorities. We propose that in order to explain these seemingly discrepant findings, these two inferences should not be treated independently. We developed an experimental paradigm to control and study these inferences simultaneously, and showed that these two inferences indeed depend, with exquisite sensitivity, on one another. Further, we proposed and validated a computational framework to explain such contrasting inferences parsimoniously, modeling observers as making rational joint inferences of wrongness and punisher’s motivation by inverting a Bayesian causal model of how authorities make punitive decisions. By characterizing how people jointly infer wrongness and legitimacy we can begin to illuminate why real world punishment attempts may fail or even backfire. contrasting inferences parsimoniously, modeling observers as making rational joint inferences of wrongness and punisher’s motivation by inverting a Bayesian causal model of how authorities make punitive decisions. By characterizing how people jointly infer wrongness and legitimacy we can begin to illuminate why real world punishment attempts may fail or even backfire.

Sara Kornfeld Simpson, Year 5 Graduate Student in Mark Bear’s Lab

Talk Title: Identifying and correcting deficits in primary visual cortex activity in Fragile X Syndrome

Abstract: Fragile X syndrome (FXS) is the most common known inherited cause of intellectual disability and autism spectrum disorder (ASD). Therapeutics for the many devastating symptoms of FXS, including deficits in sensory processing, remain elusive. Using a mouse model of FXS, we sought to discover and treat deficits in primary visual cortex (V1) activity that could be identified from a single-session of passive viewing of stimuli, with the dual aim of identifying a potential biomarker and furthering our understanding about how activity of excitatory and inhibitory circuits in V1 are changed during the disease. To interrogate this, we presented Fmr1^-/y mice and control littermates with a set of oriented, phase reversing gratings that ramped through different temporal frequencies of phase reversals ranging from 2-15 Hz, while recording the local field potential (LFP) in V1. Across all temporal frequencies, there was a deficit in response to the onset of visual stimulation in Fmr1^-/y mice. Following stimulus onset, for frequencies ranging from 4-10 Hz, there was a reduction in the evoked response of Fmr1^-/y mice. A deficit in this same theta frequency range was observed when the mice were simply viewing a static gray screen; this deficit, along with an increase in gamma power (30-85 Hz), was present in almost every Fmr1^-/y mouse, suggesting this is a useful biomarker. The difference in theta power between genotypes went away when the animals were sitting in the dark, implicating a deficit in activity of a specific sub-population of inhibitory interneurons which are activated during visual stimulation. Of particular interest, a similar decrease in power in the theta frequency band was observed when measuring EEG in posterior cortex of children with FXS. This suggests a conserved phenotype across species, underscoring the viability of this as a tool for diagnosis and testing therapeutics. 

Jennifer Hu, Year 5 Graduate Student in Roger Levy’s Lab

Talk Title: What do language models know about meaning?

Abstract:  Humans communicate not just through the literal meanings of words and sentences, but through inferential processes that give rise to rich, context-dependent meanings. For example, listeners interpret speaker utterances by considering the unspoken sentences that the speaker chose not to say, and even flout literal meanings to interpret ironic or indirect statements. In this talk, I use artificial neural network language models (NLMs) to investigate how humans comprehend language in these flexible, pragmatic ways. We find that NLMs capture many aspects of pragmatic language understanding, without any language-specific inductive biases or mental state representations. Our results suggest that domain-general prediction mechanisms support pragmatic processing, illustrating how artificial models can yield mechanistic insights into human language comprehension.