IST Austria: PubRep (Institute of Science and Technology)
Not a member yet
6140 research outputs found
Sort by
Scaling laws for sparsely-connected foundation models
We explore the impact of parameter sparsity on the scaling behavior of Transformers trained on massive datasets (i.e., "foundation models"), in both vision and language domains. In this setting, we identify the first scaling law describing the relationship between weight sparsity, number of non-zero parameters, and amount of training data, which we validate empirically across model and data scales; on ViT/JFT-4B and T5/C4. These results allow us to characterize the "optimal sparsity", the sparsity level which yields the best performance for a given effective model size and training budget. For a fixed number of non-zero parameters, we identify that the optimal sparsity increases with the amount of data used for training. We also extend our study to different sparsity structures (such as the hardware-friendly n:m pattern) and strategies (such as starting from a pretrained dense model). Our findings shed light on the power and limitations of weight sparsity across various parameter and computational settings, offering both theoretical understanding and practical implications for leveraging sparsity towards computational efficiency improvements. We provide pruning and scaling law fitting code at: github.com/google-research/jaxpruner/tree/main/jaxpruner/projects/bigsparse
Chromosomal inversion polymorphisms are widespread across the species ranges of rough periwinkles (Littorina saxatilis and L. arcana)
Inversions are thought to play a key role in adaptation and speciation, suppressing recombination between diverging populations. Genes influencing adaptive traits cluster in inversions, and changes in inversion frequencies are associated with environmental differences. However, in many organisms, it is unclear if inversions are geographically and taxonomically widespread. The intertidal snail, Littorina saxatilis, is one such example. Strong associations between putative polymorphic inversions and phenotypic differences have been demonstrated between two ecotypes of L. saxatilis in Sweden and inferred elsewhere, but no direct evidence for inversion polymorphism currently exists across the species range. Using whole genome data from 107 snails, most inversion polymorphisms were found to be widespread across the species range. The frequencies of some inversion arrangements were significantly different among ecotypes, suggesting a parallel adaptive role. Many inversions were also polymorphic in the sister species, L. arcana, hinting at an ancient origin
A structurally precise mechanism links an epilepsy-associated KCNC2 potassium channel mutation to interneuron dysfunction
De novo heterozygous variants in KCNC2 encoding the voltage-gated potassium (K+) channel subunit Kv3.2 are a recently described cause of developmental and epileptic encephalopathy (DEE). A de novo variant in KCNC2 c.374G > A (p.Cys125Tyr) was identified via exome sequencing in a patient with DEE. Relative to wild-type Kv3.2, Kv3.2-p.Cys125Tyr induces K+ currents exhibiting a large hyperpolarizing shift in the voltage dependence of activation, accelerated activation, and delayed deactivation consistent with a relative stabilization of the open conformation, along with increased current density. Leveraging the cryogenic electron microscopy (cryo-EM) structure of Kv3.1, molecular dynamic simulations suggest that a strong π-π stacking interaction between the variant Tyr125 and Tyr156 in the α-6 helix of the T1 domain promotes a relative stabilization of the open conformation of the channel, which underlies the observed gain of function. A multicompartment computational model of a Kv3-expressing parvalbumin-positive cerebral cortex fast-spiking γ-aminobutyric acidergic (GABAergic) interneuron (PV-IN) demonstrates how the Kv3.2-Cys125Tyr variant impairs neuronal excitability and dysregulates inhibition in cerebral cortex circuits to explain the resulting epilepsy
Too much information? Males convey parasite levels using more signal modalities than females utilise
Elaborate sexual signals are thought to have evolved and be maintained to serve as honest indicators of signaller quality. One measure of quality is health, which can be affected by parasite infection. Cnemaspis mysoriensis is a diurnal gecko that is often infested with ectoparasites in the wild, and males of this species express visual (coloured gular patches) and chemical (femoral gland secretions) traits that receivers could assess during social interactions. In this paper, we tested whether ectoparasites affect individual health, and whether signal quality is an indicator of ectoparasite levels. In wild lizards, we found that ectoparasite level was negatively correlated with body condition in both sexes. Moreover, some characteristics of both visual and chemical traits in males were strongly associated with ectoparasite levels. Specifically, males with higher ectoparasite levels had yellow gular patches with lower brightness and chroma, and chemical secretions with a lower proportion of aromatic compounds. We then determined whether ectoparasite levels in males influence female behaviour. Using sequential choice trials, wherein females were provided with either the visual or the chemical signals of wild-caught males that varied in ectoparasite level, we found that only chemical secretions evoked an elevated female response towards less parasitised males. Simultaneous choice trials in which females were exposed to the chemical secretions from males that varied in parasite level further confirmed a preference for males with lower parasites loads. Overall, we find that although health (body condition) or ectoparasite load can be honestly advertised through multiple modalities, the parasite-mediated female response is exclusively driven by chemical signals.</jats:p
Zigzag optical cavity for sensing and controlling torsional motion
Precision sensing and manipulation of milligram-scale mechanical oscillators has attracted growing interest in the fields of table-top explorations of gravity and tests of quantum mechanics at macroscopic scales. Torsional oscillators present an opportunity in this regard due to their remarked isolation from environmental noise. For torsional motion, an effective employment of optical cavities to enhance optomechanical interactions—as already established for linear oscillators—so far faced certain challenges. Here, we propose a concept for sensing and manipulating torsional motion, where exclusively the torsional rotations of a pendulum are mapped onto the path length of a single two-mirror optical cavity. The concept inherently alleviates many limitations of previous approaches. A proof-of-principle experiment is conducted with a rigidly controlled pendulum to explore the sensing aspects of the concept and to identify practical limitations in a potential state-of-the art setup. Based on this study, we anticipate development of precision torque sensors utilizing torsional pendulums that can support sensitivities below 10−19Nm/√Hz, while the motion of the pendulums are dominated by quantum radiation pressure noise at sub-microwatts of incoming laser power. These developments will provide horizons for experiments at the interface of quantum mechanics and gravity
LIPIcs
Traditional blockchains grant the miner of a block full control not only over which transactions but also their order. This constitutes a major flaw discovered with the introduction of decentralized finance and allows miners to perform MEV attacks. In this paper, we address the issue of sandwich attacks by providing a construction that takes as input a blockchain protocol and outputs a new blockchain protocol with the same security but in which sandwich attacks are not profitable. Furthermore, our protocol is fully decentralized with no trusted third parties or heavy cryptography primitives and carries a linear increase in latency and minimum computation overhead
Nanocrystal assemblies: Current advances and open problems
We explore the potential of nanocrystals (a term used equivalently to nanoparticles) as building blocks for nanomaterials, and the current advances and open challenges for fundamental science developments and applications. Nanocrystal assemblies are inherently multiscale, and the generation of revolutionary material properties requires a precise understanding of the relationship between structure and function, the former being determined by classical effects and the latter often by quantum effects. With an emphasis on theory and computation, we discuss challenges that hamper current assembly strategies and to what extent nanocrystal assemblies represent thermodynamic equilibrium or kinetically trapped metastable states. We also examine dynamic effects and optimization of assembly protocols. Finally, we discuss promising material functions and examples of their realization with nanocrystal assemblies
Protocol for sorting cells from mouse brains labeled with mosaic analysis with double markers by flow cytometry
Mosaic analysis with double markers (MADM) technology enables the generation of genetic mosaic tissue in mice and high-resolution phenotyping at the individual cell level. Here, we present a protocol for isolating MADM-labeled cells with high yield for downstream molecular analyses using fluorescence-activated cell sorting (FACS). We describe steps for generating MADM-labeled mice, perfusion, single-cell suspension, and debris removal. We then detail procedures for cell sorting by FACS and downstream analysis. This protocol is suitable for embryonic to adult mice.
For complete details on the use and execution of this protocol, please refer to Contreras et al. (2021).
Noncommutative Bohnenblust–Hille inequalities
Bohnenblust–Hille inequalities for Boolean cubes have been proven with dimension-free constants that grow subexponentially in the degree (Defant et al. in Math Ann 374(1):653–680, 2019). Such inequalities have found great applications in learning low-degree Boolean functions (Eskenazis and Ivanisvili in Proceedings of the 54th annual ACM SIGACT symposium on theory of computing, pp 203–207, 2022). Motivated by learning quantum observables, a qubit analogue of Bohnenblust–Hille inequality for Boolean cubes was recently conjectured in Rouzé et al. (Quantum Talagrand, KKL and Friedgut’s theorems and the learnability of quantum Boolean functions, 2022. arXiv preprint arXiv:2209.07279). The conjecture was resolved in Huang et al. (Learning to predict arbitrary quantum processes, 2022. arXiv preprint arXiv:2210.14894). In this paper, we give a new proof of these Bohnenblust–Hille inequalities for qubit system with constants that are dimension-free and of exponential growth in the degree. As a consequence, we obtain a junta theorem for low-degree polynomials. Using similar ideas, we also study learning problems of low degree quantum observables and Bohr’s radius phenomenon on quantum Boolean cubes
Locally free representations of quivers over commutative Frobenius algebras
In this paper we investigate locally free representations of a quiver Q over a commutative Frobenius algebra R by arithmetic Fourier transform. When the base field is finite we prove that the number of isomorphism classes of absolutely indecomposable locally free representations of fixed rank is independent of the orientation of Q. We also prove that the number of isomorphism classes of locally free absolutely indecomposable representations of the preprojective algebra of Q over R equals the number of isomorphism classes of locally free absolutely indecomposable representations of Q over R[t]/(t2). Using these results together with results of Geiss, Leclerc and Schröer we give, when k is algebraically closed, a classification of pairs (Q, R) such that the set of isomorphism classes of indecomposable locally free representations of Q over R is finite. Finally when the representation is free of rank 1 at each vertex of Q, we study the function that counts the number of isomorphism classes of absolutely indecomposable locally free representations of Q over the Frobenius algebra Fq[t]/(tr). We prove that they are polynomial in q and their generating function is rational and satisfies a functional equation