Integrated Calendar

A defining challenge in synthetic biology is the development of high-throughput and automated methodologies for precise design and engineering of cells. To address these challenges, we develop multiplex genome engineering technologies for versatile genome modification and evolution of bacterial and eukaryotic cells. We use these technologies to create genetic variants to reveal a causal understanding of complex phenotypes as well as engineer pathways and recode genomes. These Genomically Recoded Organisms (GROs) contain alternative genetic codes, in which codons have been eliminated from the genome of E. coli. GROs exhibit improved properties for incorporation of nonstandard amino acids that expand the chemical diversity of proteins or polymers, establish genetic isolation and multi-virus resistance, and enable the engineering of GROs to depend on synthetic amino acids for robust biocontainment strategies. We have also developed new computational-experimental technologies – computer aided design of synthetic genetic elements (CAD-SGE) – that permits the redesign, expression, and mobilization of biosynthetic pathways in diverse organisms for the discovery of new metabolites. This work increases the toolbox for genomic and cellular engineering with broad applications for new classes of enzymes, materials, and therapeutics.

In this talk, I reverse engineer CLIP, one of the most commonly used computer vision backbones. I analyze how individual model components affect the final CLIP representation. I show that the image representation can be decomposed as a sum across individual image patches, model layers, neurons, and attention heads, and use CLIP’s text representation to interpret the summands.

When interpreting the attention heads, each head role can be characterized by automatically finding text representations that span its output space, which reveals property-specific roles for many heads (e.g. location or shape). Next, interpreting the image patches uncovers an emergent spatial localization within CLIP. Finally, the automatic description of the contributions of individual neurons shows polysemantic behavior - each neuron corresponds to multiple, often unrelated, concepts (e.g. ships and cars).

The gained understanding of different components allows three main applications: First, the discovered head roles enable the removal of spurious features from CLIP. Second, emergent localization is used for a strong zero-shot image segmenter. Finally, the extracted neuron polysemy allows the mass production of “semantic” adversarial examples by generating images with concepts spuriously correlated to the incorrect class. The results indicate that a scalable understanding of transformer models is attainable and can be used to detect model bugs, repair them, and improve them.  

BIO:

Yossi is a computer science PhD at UC Berkeley, advised by Alexei Efros, and a visiting researcher at Meta. Before that, he was a member of the perception team at Google Research (now Google-DeepMind). He completed his M.Sc. at Weizmann Institute, advised by Prof. Michal Irani. His research centers around deep learning, computer vision, and mechanistic interpretability.

The role of neurons in the direction-selective retinal circuit in visual processing in the retina and in the visual thalamus

The lateral geniculate nucleus (LGN) of the thalamus is a major retinal target, involved in processing and relaying visual information, including direction selectivity (DS) and orientation selectivity (OS). How DS and OS are organized in the LGN is poorly understood, as well as whether this information is directly inherited from the retina or generated de novo within the LGN. Using extracellular recordings from across the mouse LGN, we studied DS and OS responses and their topographic organization. We found that DS responses are absent in the central visual field, and that their preferred directions are topographically aligned to match translational optic flow patterns in the remaining visual field. OS responses were uniformly distributed throughout the visual field. By eliminating retinal DS in transgenic mice, we found that DS- but not OS-responses in the LGN were dependent on retinal DS. Thus, LGN DS is inherited from the retina, but retinogeniculate transfer may be topography-dependent, optimizing representations that support visually-guided behaviors.

Pain is a self-preservation mechanism, providing warning indicators associated with tissue damage. These indicators are perceived by nociceptive peripheral innervations with the ability to signal the brain. Nociceptive innervations are also a part of the infrastructure of various organs, yet the imprint their activity has on tissue physiology remains understudied. Here, we applied chemogenetics in mice to locally activate cutaneous TRPV1 innervations in naïve skin and found it triggered accelerated anagen onset. This was preceded by a rapid apoptosis of dermal macrophages mediated by neuropeptide calcitonin gene-related peptide (CGRP), followed by an induction of Osteopontin (Spp1)-expressing dermal fibroblasts. Spp1, an extracellular matrix protein and a hair growth promoting factor, was essential for the TRPV1-triggered induction of new regenerative cycling by dormant hair follicles. Specifically, macrophages responsiveness to CGRP was required for the changes in dermal fibroblasts. Finally, we show that epidermal abrasion induced Spp1-expressing dermal fibroblasts and hair growth via a TRPV1 neuron and CGRP dependent mechanism. Collectively, these data demonstrate a role for pain facilitating innervations in coordinating a cellular mechanism that promotes hair growth and the restoration of this important mechano- and thermo-protective barrier

Seismic waves excited by human activity frequently obscure signals
due to tectonic processes and are discarded as a nuisance. Seismic
noise-field analysis is, however, a powerful tool for characterizing
anthropogenic activities. In this talk, I will briefly review the
seismological fingerprints of anthropogenic noise sources and then
present a scheme devised to identify precursory activity leading to the
October 7 terrorist attack. The precursory activity in Gaza included
massive mobilization, documented by multiple media outlets. Favorable
conditions arose due to a temporary lack of anthropogenic activity in
Israel, allowing remote seismic stations to record signals due to Gaza
vehicle traffic in the early hours of Oct. 7. Seismogram analysis reveals
a widespread signal that abruptly emerged above the nighttime noise
levels about 20 minutes before the attack began. Statistical analysis
suggests the signal is highly anomalous; tests for significance indicate
that pre-attack inter-station correlations would emerge by chance only
once every 18,000 years. Tripartite array analysis was used to detect
surface waves, locate their sources, and demarcate the extent of preattack
activity within the Gaza Strip. The signal’s amplitude, frequency,
and spatiotemporal distribution appear to be aligned with vehicular
traffic emanating from the south-central region of the Gaza Strip and
extending towards its peripheries in the half-hour window preceding the
invasion. This provides valuable tactical information and suggests
embedding seismic noise-field analysis into decision-making protocols
could enhance preparedness for terrorist attacks.