Harvard University Department of Physics

Science Research Public Lecture: Jelani Nelson

Sketching Big Data
Wed, Oct 25 2017 @7:00pm
Science Center Hall C
One Oxford St., Cambridge, MA

How to Grow a Spine

Like a string of pearls, the spine is made of a series of similar vertebrae. A so-called segmentation clock creates this repetitive arrangement in developing embryos: Each time the clock ticks, a vertebra starts to form...

Toward Living Matter with Colloidal Particles

Understanding the differences between inanimate matter and living matter is one of the fundamental unsolved problems of science. Although this question might be framed as philosophical...

UPCOMING EVENTS

Monday, October 23rd

Monday Colloquium: Srini Turaga (HHMI Janelia), "From biological neural networks to artificial neural networks"
Start: 04:15pm - End: 05:30pm
- 04:15pm
- In this talk, I will describe how we developed deep learning based computational tools to solve two important problems in neuroscience: predicting the activity of a neural network from measurements of its structural connectivity, and inferring the connectivity of a network of neurons from measurements and perturbation of neural activity.
  
  1. Are measurements of the structural connectivity of a biological neural network sufficient to predict its function? We constructed a simplified model of the first two stages of the fruit fly visual system, the lamina and medulla. The result is a deep hexagonal lattice convolutional neural network which discovered well-known orientation and direction selectivity properties in T4 neurons and their inputs. Our work is the first demonstration, that knowledge of neural connectivity can enable in silico predictions of the functional properties of individual neurons in a circuit, leading to an understanding of circuit function from structure alone.
  
  2. Can we infer neural connectivity from noisy measurement and perturbation of neural activity? Population neural activity measurement by calcium imaging can be combined with cellular resolution optogenetic activity perturbations to enable the mapping of neural connectivity in vivo. This requires accurate inference of perturbed and unperturbed neural activity from calcium imaging measurements, which are noisy and indirect. We built on recent advances in variational autoencoderes to develop a new fully Bayesian approach to jointly inferring spiking activity and neural connectivity from in vivo all-optical perturbation experiments. Our model produces excellent spike inferences at 20K times real-time, and predicts connectivity for mouse primary visual cortex which is consistent with known measurements.
- View in Google Calendar

Wednesday, October 25th

Prof. Mike Hagan - Brandeis University
Start: 05:30pm - End: 07:00pm
- 05:30pm
- View in Google Calendar

Thursday, October 26th

CMP Seminar: Masaki Oshikawa, University of Tokyo
Start: 12:00pm - End: 01:00pm
- 12:00pm
- View in Google Calendar

Monday, October 30th

Monday Colloquium: Wolf Rueckner (Harvard), Title TBA
Start: 04:15pm - End: 05:30pm
- 04:15pm
- View in Google Calendar

Wednesday, November 1st

Dr. Yaniv Edery - Harvard University
Start: 05:30pm - End: 07:00pm
- 05:30pm
- View in Google Calendar