Pages
June 01, 2015
-
Date:07ThursdayDecember 2023Lecture
Spatial Biology by Imaging Mass Cytometry
More information Time 09:00 - 10:00Location ZOOMLecturer Dr. Sean Pawlowski(Ionpath) & Dr. Tomer-Meir Salame (LSCF)
Mass Cytometry UnitOrganizer Department of Life Sciences Core FacilitiesHomepage Contact -
Date:12TuesdayDecember 2023Lecture
Open Day at SAMPL Lab
More information Time 15:00 - 16:30Location Ullmann Building of Life SciencesLecturer Modeling Processing and Learning Lab., Prof. Yonina Eldar Organizer Department of MathematicsHomepage Contact -
Date:19TuesdayDecember 2023Lecture
From Randomness to Function: de novo Proteins as a Source of Molecular and Cellular Innovation
More information Time 11:00 - 12:00Location Nella and Leon Benoziyo Building for Biological SciencesLecturer Dr. Idan Frumkin
Life Sciences, Massachusetts Institute of Technology (MIT), USAOrganizer Department of Biomolecular SciencesContact Abstract Show full text abstract about How do novel genes emerge? While new genes often evolve from...» How do novel genes emerge? While new genes often evolve from older ones, novelty can also originate from random sequences in a process termed "de novo gene birth". The functions of such de novo genes and how they integrate into complex cellular systems are poorly understood. By screening a library of 100 million random proteins with no sequence similarity to existing proteins, we identified thousands of functional proteins promoting E. coli survival against a toxin or a bacterial virus. Using genetic and biochemical tools, we characterized selected random proteins and revealed they integrate into pre-existing cellular pathways to mitigate cellular threats. Our work provides a mechanistic basis for understanding how de novo gene birth can yield functional proteins that effectively benefit cells evolving under stress.
-
Date:20WednesdayDecember 2023Lecture
"Shedding light on the dark matter of viral proteomes to advance our understanding of antiviral immunity"
More information Time 10:00 - 12:00Location Nella and Leon Benoziyo Building for Biological SciencesLecturer Dr. Shira Weingarten-Gabbay
Broad Institute of MIT and HarvardOrganizer Department of Molecular Cell BiologyContact -
Date:21ThursdayDecember 2023Lecture
Geometric Functional Analysis and Probability Seminar
More information Time 13:30 - 14:30Title Gaps of Fourier Quasicrystals and Lee-Yang PolynomialsLocation Jacob Ziskind BuildingLecturer Lior Alon
MITOrganizer Department of MathematicsContact Abstract Show full text abstract about The concept of "quasi-periodic" sets, functions, a...» The concept of "quasi-periodic" sets, functions, and measures is
prevalent in diverse mathematical fields such as Mathematical Physics,
Fourier Analysis, and Number Theory. The Poisson summation formula provides a “Fourier characterization” for discrete periodic sets, saying that the Fourier transform of the counting measure of a discrete periodic set is also a counting measure of a discrete periodic set. Fourier Quasicrystals (FQ) generalize this notion of periodicity: a counting measure of a discrete set is called a Fourier quasicrystal (FQ) if its Fourier transform is also a discrete atomic measure, together with some growth condition.
Recently Kurasov and Sarnak provided a method for construction of one-dimensional counting measures which are FQ (motivated by quantum graphs) using the torus zero sets of multivariate Lee-Yang polynomials. In this talk, I will show that the Kurasov-Sarnak construction generates all FQ counting measures in 1D.
A discrete set on the real line is fully described by the gaps between consecutive points. A discrete periodic set has finitely many gaps. We show that a non-periodic FQ has uncountably many gaps, with a well-defined gap distribution. This distribution is given explicitly in terms of an ergodic dynamical system induced from irrational flow on the torus.
The talk is aimed at a broad audience, no prior knowledge in the field is assumed.
Based on joint works with Alex Cohen and Cynthia Vinzant.
-
Date:26TuesdayDecember 2023Lecture
The structure of protein complexes underlies co-translational assembly
More information Time 14:00 - 15:00Location Gerhard M.J. Schmidt Lecture HallOrganizer Department of Chemical and Structural BiologyContact -
Date:27WednesdayDecember 2023Lecture
Membrane dynamics during giant vesicle secretion
More information Time 10:00 - 11:00Location Arthur and Rochelle Belfer Building for Biomedical ResearchLecturer Prof. Benny Shilo
Dept of Molecular GeneticsContact -
Date:27WednesdayDecember 2023Lecture
Foundations of Computer Science Seminar
More information Time 11:15 - 12:15Title Explicit Codes for Poly-Size Circuits and Functions that are Hard to Sample on Low Entropy DistributionsLocation Jacob Ziskind BuildingLecturer Jad Silbak
Northeastern UniversityOrganizer Department of Computer Science and Applied MathematicsContact Abstract Show full text abstract about Guruswami and Smith (J. ACM 2016) considered codes for chann...» Guruswami and Smith (J. ACM 2016) considered codes for channels that are computationally bounded which modify at most a p-fraction of the bits of the codeword. This class of channels is significantly stronger than Shannon’s binary symmetric channel which flips each bit independently with probability p, but weaker than Hamming’s channels which may flip any p-fraction of bits and are computationally unbounded.
Recently, there has been a growing body of work that aims to construct codes against channels that are computationally bounded (e.g., bounded memory channels, channels that are poly-size circuits). In this work, we consider bounded size channels and construct codes that:
- Achieve an optimal rate of 1-H(p) (matching the rate of binary symmetric channels, and beating the rate of Hamming channels).
- Are explicit, assuming E does not have a size 2^Ω(n) nondeterministic circuits.
Achieving these codes implies circuit lower bounds (and therefore explicit constructions need to be based on hardness assumptions). This result builds on the recent result by Shaltiel and Silbak (FOCS 2022) that gave a randomized Monte-Carlo construction, rather than explicit codes.
A key component in our codes (that we believe to be of independent interest) is a new complexity theoretic notion of hard to sample functions (HTS).
Loosely speaking, a function f is HTS for circuits of size n^c, if for every randomized circuit A of size n^c that samples a distribution (X,Y) such that X has sufficiently large min-entropy, it holds that Pr[Y=f(X)] is small.
This notion is inspired by a related notion introduced by Viola (SICOMP 2012) in which X is the uniform distribution and is similar in flavor to the definition of multi-collision-resistant hash functions. Building on classical works on “hardness amplification” (and using many additional tools and ideas from pseudorandomness) we construct HTS functions under hardness assumptions.
This is a joint work with Ronen Shaltiel.
-
Date:27WednesdayDecember 2023Lecture
LS Luncheon
More information Time 12:00 - 14:00Title The tumor micro(b)environment and its effects on response to therapyLocation Nella and Leon Benoziyo Building for Biological SciencesLecturer Prof. Ravid Straussman
Dept of Molecular Cell BiologyContact -
Date:28ThursdayDecember 2023Lecture
Vision and AI
More information Time 12:15 - 13:15Title A theory of Unsupervised Translation Motivated by Understanding whale communicationLocation Jacob Ziskind BuildingLecturer Shafi Goldwasser
Simons Institute, UC BerkeleyOrganizer Department of Computer Science and Applied MathematicsContact Abstract Show full text abstract about Neural networks are capable of translating between languages...» Neural networks are capable of translating between languages—in some cases even between two languages where there is little or no access to parallel translations, in what is known as Unsupervised Machine Translation (UMT). Given this progress, it is intriguing to ask whether machine learning tools can ultimately enable understanding animal communication, particularly that of highly intelligent animals. We propose a theoretical framework for analyzing UMT when no parallel translations are available and when it cannot be assumed that the source and target corpora address related subject domains or posses similar linguistic structure.
-
Date:28ThursdayDecember 2023Lecture
Geometric Functional Analysis and Probability Seminar
More information Time 13:30 - 14:30Title Persistence of stationary Gaussian fields with spectral singularity: from universality to entropic repulsionLocation Jacob Ziskind BuildingLecturer Ohad Noy Feldheim
HUJIOrganizer Department of MathematicsContact Abstract Show full text abstract about We study the persistence event ${f(x)>0 for all x in D}$ ...» We study the persistence event ${f(x)>0 for all x in D}$ for stationary Gaussian fields on d dimensional Euclidean space, with a spectral singularity at the origin of order $alpha$, i.e. such that their spectral density at the origin behaves like $|lambda|^{-alpha}$. Our main result is that, under certain mild regularity conditions, the probability of persistence on a large ball $D=B(T)$ decays at a universal log-asymptotic rate of $m (d-alpha)(log T) capa(B(T))$, where $capa(B(T))$ is the capacity of $B(T)$ with respect to the field, and $m$ is the mass of the absolutely continuous component of its spectrum. This generalises a result of Bolthausen, Deuschel and Zeitouni for the Gaussian free field (GFF) on $^d$, $d ge 3$, to a wide class of Gaussian fields with spectral singularity. Using this estimate, we establish an `entropic repulsion’ phenomenon, showing that the field conditioned to persist is tending in distribution to an independent copy of the unconditioned process, shifted by a constant level of $sqrt{m (d-alpha)(log T)}$.
This work is the first to introduce the notions of capacity and equilibrium potential to the spectral theory of persistence. In the course of the talk I will explain these notions, provide an overview of the proof and present the role of these notions in improving the precision of persistence probability estimates. I will also try to put this work in the context of current study of rare events of stationary Gaussian processes. No prior knowledge of the subject is assumed.
Joint work with Naomi Feldheim from Bar-Ilan and Stephen Muirhead from Melbourne.
-
Date:31SundayDecember 2023Lecture
The Clore Center for Biological Physics
More information Time 13:15 - 14:15Title Erasing information fast and cheap --- How to approach Landauer’s bound?Location Edna and K.B. Weissman Building of Physical SciencesOrganizer Clore Center for Biological PhysicsContact Abstract Show full text abstract about The celebrated Landauer bound is the fundamental universal c...» The celebrated Landauer bound is the fundamental universal cost of computation: there must be dissipation of at least kBTlog2 per erasure of one bit. This fundamental bound is reached when the erasure protocol is performed in the slow quasi-static limit. Generally, the faster the erasure protocol, the more dissipation is generated.
In this talk, I will present two approaches that challenge this view. First, it will be shown that by the use of a conserved quantity in the system, one can bypass Liouville’s theorem and perform erasure at zero energetic cost. The second approach that will be discussed is considering a system that is weakly coupled to the environment. In that case, one can design an erasure procedure that does not scale with its operation time.
-
Date:02TuesdayJanuary 2024Lecture
A paradigm shift in GPCR recruitment and activity: GPCR Voltage Dependence Controls Neuronal Plasticity and Behavior
More information Time All dayLocation Gerhard M.J. Schmidt Lecture HallLecturer Prof. Moshe Parnas
Dept of Physiology and Pharmacology Tel Aviv UniversityOrganizer Department of Brain SciencesContact Abstract Show full text abstract about : G-protein coupled receptors (GPCRs) play a paramount role ...» : G-protein coupled receptors (GPCRs) play a paramount role in diverse brain functions. Twenty years ago, GPCR activity was shown to be regulated by membrane potential in vitro, but whether the voltage dependence of GPCRs contributes to neuronal coding and behavioral output under physiological conditions in vivo has never been demonstrated. We show in two different processes that muscarinic GPCR mediated neuromodulation in vivo is voltage dependent. First, we show that muscarinic type A receptors (mAChR-A) mediated neuronal potentiation is voltage dependent. This potentiation voltage dependency is abolished in mutant flies expressing a voltage independent receptor. Most important, muscarinic receptor voltage independence caused a strong behavioral effect of increased odor habituation. Second, we show that muscarinic type B receptors (mAChR-B) voltage dependency is required for both efficient and accurate learning and memory. Normally, to prevent non-specific olfactory learning and memory, mAChR-B activity suppress both signals that are required for plasticity. Behavior experiments demonstrate that mAChR-B knockdown impairs olfactory learning by inducing undesired changes to the valence of an odor that was not associated with the reinforcer. On the other hand, mAChR-B voltage dependence prevents mAChR-B to interfere with plasticity in neurons that are required for the learning and memory process. Indeed, generating flies with a voltage independent mAChR-B resulted in impaired learning. Thus, we provide the very first demonstrations of physiological roles for the voltage dependency of GPCRs by demonstrating crucial involvement of GPCR voltage dependence in neuronal plasticity and behavior. As such, our findings create a paradigm shift in our thinking on GPCR recruitment and activity. Together, we suggest that GPCR voltage dependency plays a role in many diverse neuronal functions including learning and memory and may serve as a target for novel drug development.
Light refreshments before the seminar.
-
Date:02TuesdayJanuary 2024Lecture
Enhanced Growth in Atomic Layer Deposition of Transition Metals: The Role of Surface Diffusion and Nucleation Sites
More information Time 11:15 - 12:15Location Gerhard M.J. Schmidt Lecture HallLecturer Dr. Amnon Rothman
Chemical Engineering, Stanford UniversityOrganizer Department of Molecular Chemistry and Materials ScienceContact Abstract Show full text abstract about Noble metal thin films have attracted significant interest o...» Noble metal thin films have attracted significant interest owing to their distinctive properties and structures, which make them ideal for applications in microelectronics, catalysis, energy, and photovoltaics. While several parameters influence the properties of these metals for such applications, the deposition process remains a critical factor. Atomic Layer Deposition (ALD) stands out as a prevalent deposition technique due to its surface-sensitive nature. The ALD process is characterized by its self-limiting surface reactions, promoting a layer-by-layer growth mechanism and allowing for precise control over film thickness and conformality. However, challenges arise in achieving continuous, pinhole-free noble metal ALD layers on oxide surfaces, often resulting in low film quality. These challenges can be traced back to the lack of adequate nucleation sites and the poor wettability of the low-surface energy substrates. The research studied the impact of substrate surface functionalization using organometallic molecules, such as trimethylaluminum (TMA) and diethylzinc (DEZ), on the nucleation and growth of Ru layers. The results reveal an enhancement in both nucleation density and the average diameter of the Ru nanoparticles deposited, and these improvements were attributed to an increase in both nucleation sites and elevated surface diffusivity. The latter effect is speculated to result from a reduction in the substrate's surface free energy. The study also examines the influence of substrate surface characteristics, including surface termination and crystallinity, on the nucleation and growth of Ru metal via ALD. The morphologies of the resulting Ru thin films are studied using scanning electron microscopy (SEM), atomic force microscopy (AFM), and grazing incidence small angle x-ray scattering (GISAXS). These analytical results are integrated with an experimental model to elucidate the differences in growth mechanisms observed across substrates. The findings underscore the importance of substrate choice in the ALD process and broaden our understanding of Ru metal growth. This research serves as an important step in optimizing the ALD process for various applications by tailoring substrate selection. -
Date:03WednesdayJanuary 2024Lecture
Special Guest Seminar: Dr. Sharon Fleischer
More information Time 10:00 - 12:00Title Stem cell based cardiac tissue models to study the human heart in health and diseaseLocation Wolfson Building for Biological ResearchLecturer Dr. Sharon Fleischer
Columbia UniversityOrganizer Department of Molecular Cell BiologyContact -
Date:03WednesdayJanuary 2024Lecture
Machine Learning and Statistics Seminar
More information Time 11:15 - 12:15Title Theoretical Foundations of Neural Networks: Expressiveness, Optimization, and GeneralizationLocation Jacob Ziskind BuildingOrganizer Department of Computer Science and Applied MathematicsContact Abstract Show full text abstract about In recent years neural networks achieved state-of-the-art re...» In recent years neural networks achieved state-of-the-art results in many different applications such as computer vision, natural language processing, and speech recognition. Despite their empirical success, the theoretical understanding of neural networks remains limited. This gap can be seen in all aspects of learning: optimization, generalization, and expressiveness. My thesis aims to address this gap by providing theoretical insights into neural network success, bridging the practical and theoretical aspects. In this talk, I focus on the three following topics:
(1) Expressiveness: We study the memorization capabilities of neural networks, and provide a construction for memorization of $N$ labeled samples using a network with $ ilde{O}(sqrt{N})$ parameters, which is optimal up to log factors due to known lower bounds. Our analysis underscores the crucial role of depth in achieving near-optimal memorization capacity.
(2) Optimization: Investigating the effects of mild over-parameterization on the optimization landscape of a simple ReLU network in a well-studied teacher-student setting. We prove that the optimization landscape changes significantly between exact and over-parameterized regimes. For example, over-parameterization can turn local minima into saddle points in certain cases, making the landscape more favorable.
(3) Generalization: Exploring the generalization properties of interpolating 2-layer neural networks in the presence of label noise. We prove that in this overfitting regime, the input dimension plays a major role in the type of overfitting the network achieves. Namely, in dimension 1 the overfitting scales with the magnitude of the noise (tempered overfitting), while in high dimensions it converges to zero with the input dimension (benign overfitting).
-
Date:03WednesdayJanuary 2024Lecture
Chemical and Biological Physics Guest seminar
More information Time 15:00 - 16:00Title Atomic arrays as programmable quantum processors and sensorsLocation Gerhard M.J. Schmidt Lecture HallLecturer Dr Ran Finkelstein
CaltechOrganizer Department of Chemical and Biological PhysicsContact Abstract Show full text abstract about Large arrays of trapped neutral atoms have emerged over the ...» Large arrays of trapped neutral atoms have emerged over the past few years as a promising platform for quantum information processing, combining inherent scalability with high-fidelity control and site-resolved readout. In this talk, I will discuss ongoing work with arrays of Alkaline-earth atoms. These divalent atoms offer unique properties stemming largely from their long-lived metastable states, which form the basis of the optical atomic clock. I will describe the design of a universal quantum processor based on clock qubits and its application in quantum metrology, and I will address the challenge of generating and benchmarking highly entangled states in an analog quantum simulator. First, we realize scalable local control of individual clock qubits, which we utilize to extend the Ramsey interrogation time beyond the coherence time of a single atom [1]. To realize a universal quantum processor, we demonstrate record high-fidelity two-qubit entangling gates mediated by Rydberg interactions, which we combine with dynamical reconfiguration to entangle clock probes in GHZ states and perform Ancilla-based detection [2]. We then use the narrow clock transition to measure and remove thermal excitations of atoms in tweezers (a technique known as erasure conversion) and generate hyperentangled states of motion and spin [3]. In the second part of the talk, I will describe a different approach for generating large scale entangled states in an analog quantum simulator configuration [4], including error mitigation [5] and benchmarking of a 60-atom simulator [6]. Together, these show the great promise and the large variety of experiments accessed with this emerging platform. [1] A. Shaw*, R. Finkelstein*, R. Tsai, P. Scholl, T. Yoon, J. Choi, M. Endres, Multi-ensemble metrology by programming local rotations with atom movements, arxiv:2303.16885, Nature Physics in press (2023). [2] R. Finkelstein, R. Tsai, A. Shaw, X. Sun, M. Endres, A universal quantum processor for entanglement enhanced optical tweezer clocks, in preparation. [3] P. Scholl*, A. Shaw*, R. Finkelstein*, R. Tsai, J. Choi, M. Endres, Erasure cooling, control, and hyper-entanglement of motion in optical tweezers, arXiv:2311.15580 (2023). [4] J. Choi, A. Shaw, I. Madjarov, X. Xie, R. Finkelstein, J. Covey, J. Cotler, D. Mark, H.Y. Huang, A. Kale, H. Pichler, F. Brandão, S. Choi, and M. Endres, Preparing random states and benchmarking with many-body quantum chaos, Nature 617 (2023) [5] P. Scholl, A. Shaw, R. Tsai, R. Finkelstein, J. Choi, M. Endres, Erasure conversion in a high-fidelity Rydberg quantum simulator, Nature 622 (2023). [6] A. Shaw, Z. Chen, J. Choi, D.K. Mark, P. Scholl, R. Finkelstein, A. Elben, S. Choi, M. Endres, Benchmarking highly entangled states on a 60-atom analog quantum simulator, arXiv:2308.07914 (2023). -
Date:04ThursdayJanuary 2024Lecture
Advancing MRI: Sequences and Applications
More information Time 09:00 - 10:00Location Max and Lillian Candiotty BuildingLecturer Dr. Edna Furman-Haran
MRI UnitOrganizer Department of Life Sciences Core FacilitiesContact -
Date:04ThursdayJanuary 2024Lecture
Special Guest Seminar with Prof. Yegor Bazykin
More information Time 10:00 - 11:00Title “Towards understanding and forecasting evolution of pathogenic viruses”Location Arthur and Rochelle Belfer Building for Biomedical ResearchLecturer Prof. Yegor Bazykin Organizer Department of Molecular GeneticsContact -
Date:04ThursdayJanuary 2024Lecture
Vision and AI
More information Time 12:15 - 13:15Title Generative Models for AbstractionsLocation Jacob Ziskind BuildingLecturer Yael Vinker
Tel Aviv UniversityOrganizer Department of Computer Science and Applied MathematicsContact Abstract Show full text abstract about Humans often express, organize, and develop their thoughts v...» Humans often express, organize, and develop their thoughts visually, employing abstract representations like sketches, scribbles, and symbols. These abstract representations range from simple drawings made by kids to structured charts and diagrams made by researchers. Creating such abstract representations comes naturally to humans
