Search

Open-ended questions are a favored tool among instructors for assessing student understanding and encouraging critical exploration of course material. Providing feedback for such responses is a time-consuming task that can lead to overwhelmed instructors and decreased feedback quality. Many instructors resort to simpler question formats, like multiple-choice questions, which provide immediate feedback but at the expense of personalized and insightful comments. Here, we present a tool that uses large language models (LLMs), guided by instructor-defined criteria, to automate responses to open-ended questions. Our tool delivers rapid personalized feedback, enabling students to quickly test their knowledge and identify areas for improvement. We provide open-source reference implementations both as a web application and as a Jupyter Notebook widget that can be used with instructional coding or math notebooks. With instructor guidance, LLMs hold promise to enhance student learning outcomes and elevate instructional methodologies.

Deep neural networks implement a sequence of layer-by-layer operations that are each relatively easy to understand, but the resulting overall computation is generally difficult to understand. We consider a simple hypothesis for interpreting the layer-by-layer construction of useful representations: perhaps the role of each layer is to reformat information to reduce the "distance" to the desired outputs. With this framework, the layer-wise computation implemented by a deep neural network can be viewed as a path through a high-dimensional representation space. We formalize this intuitive idea of a "path" by leveraging recent advances in *metric* representational similarity. We extend existing representational distance methods by computing geodesics, angles, and projections of representations, going beyond mere layer distances. We then demonstrate these tools by visualizing and comparing the paths taken by ResNet and VGG architectures on CIFAR-10. We conclude by sketching additional ways that this kind of representational geometry can be used to understand and interpret network training, and to describe novel kinds of similarities between different models., Comment: 10 pages, submitted to ICLR 2023

Optical atomic clocks are the most accurate measurement devices ever constructed and have found many applications in fundamental science and technology. The use of highly charged ions (HCI) as a new class of references for highest accuracy clocks and precision tests of fundamental physics has long been motivated by their extreme atomic properties and reduced sensitivity to perturbations from external electric and magnetic fields compared to singly charged ions or neutral atoms. Here we present the first realisation of this new class of clocks, based on an optical magnetic-dipole transition in Ar$^{13+}$. Its comprehensively evaluated systematic frequency uncertainty of $2.2\times10^{-17}$ is comparable to that of many optical clocks in operation. From clock comparisons we improve by eight and nine orders of magnitude upon the uncertainties for the absolute transition frequency and isotope shift ($^{40}$Ar vs. $^{36}$Ar), respectively. These measurements allow us to probe the largely unexplored quantum electrodynamic nuclear recoil, presented as part of improved calculations of the isotope shift which reduce the uncertainty of previous theory by a factor of three. This work establishes forbidden optical transitions in HCI as references for cutting-edge optical clocks and future high-sensitivity searches for physics beyond the standard model., Comment: Main: 21 pages, 3 figures. Supplement: 20 pages, 2 figures. Accepted version

It has been hypothesized that some form of "modular" structure in artificial neural networks should be useful for learning, compositionality, and generalization. However, defining and quantifying modularity remains an open problem. We cast the problem of detecting functional modules into the problem of detecting clusters of similar-functioning units. This begs the question of what makes two units functionally similar. For this, we consider two broad families of methods: those that define similarity based on how units respond to structured variations in inputs ("upstream"), and those based on how variations in hidden unit activations affect outputs ("downstream"). We conduct an empirical study quantifying modularity of hidden layer representations of simple feedforward, fully connected networks, across a range of hyperparameters. For each model, we quantify pairwise associations between hidden units in each layer using a variety of both upstream and downstream measures, then cluster them by maximizing their "modularity score" using established tools from network science. We find two surprising results: first, dropout dramatically increased modularity, while other forms of weight regularization had more modest effects. Second, although we observe that there is usually good agreement about clusters within both upstream methods and downstream methods, there is little agreement about the cluster assignments across these two families of methods. This has important implications for representation-learning, as it suggests that finding modular representations that reflect structure in inputs (e.g. disentanglement) may be a distinct goal from learning modular representations that reflect structure in outputs (e.g. compositionality)., Comment: 12 main text pages, 4 main figures, 5 supplemental figures. Will be submitted to TMLR

Sampling and Variational Inference (VI) are two large families of methods for approximate inference that have complementary strengths. Sampling methods excel at approximating arbitrary probability distributions, but can be inefficient. VI methods are efficient, but may misrepresent the true distribution. Here, we develop a general framework where approximations are stochastic mixtures of simple component distributions. Both sampling and VI can be seen as special cases: in sampling, each mixture component is a delta-function and is chosen stochastically, while in standard VI a single component is chosen to minimize divergence. We derive a practical method that interpolates between sampling and VI by solving an optimization problem over a mixing distribution. Intermediate inference methods then arise by varying a single parameter. Our method provably improves on sampling (reducing variance) and on VI (reducing bias+variance despite increasing variance). We demonstrate our method's bias/variance trade-off in practice on reference problems, and we compare outcomes to commonly used sampling and VI methods. This work takes a step towards a highly flexible yet simple family of inference methods that combines the complementary strengths of sampling and VI., Comment: 9 pages, 4 figures. Submitted to UAI 2022; under double-blind review. Code available at https://github.com/wrongu/sampling-variational-demos

We report direct measurements of the frequency ratio of the 642 THz ${}^2S_{1/2} (F=0)$--${}^2F_{7/2} (F=3)$ electric octupole transition in ${}^{171}\mathrm{Yb}^+$ and the 429 THz ${}^1S_0$--${}^3P_0$ transition in ${}^{87}\mathrm{Sr}$. A series of 107 measurements has been performed at the Physikalisch-Technische Bundesanstalt between December 2012 and October 2019. Long-term variations of the ratio are larger than expected from the individual measurement uncertainties of few $10^{-17}$. The cause of these variations remains unknown. Even taking these into account, we find a fractional uncertainty of the frequency ratio of $2.5 \times 10^{-17}$, which improves upon previous knowledge by one order of magnitude. The average frequency ratio is $\nu_{\mathrm{Yb}^+} / \nu_{\mathrm{Sr}} = 1.495\,991\,618\,544\,900\,537(38)$. This represents one of the most accurate measurements between two different atomic species to date., Comment: 9 pages, 4 figures, 1 table

We report on the first coherent excitation of the highly forbidden $^2S_{1/2}\rightarrow{}^2F_{7/2}$ electric octupole (E3) transition in a single trapped ${}^{172}$Yb$^+$ ion, an isotope without nuclear spin. Using the transition in ${}^{171}$Yb$^+$ as a reference, we determine the transition frequency to be $642\,116\,784\,950\,887.6(2.4)\,$Hz. We map out the magnetic field environment using the forbidden $^2S_{1/2} \rightarrow{}^2D_{5/2}$ electric quadrupole (E2) transition and determine its frequency to be $729\,476\,867\,027\,206.8(4.4)\,$Hz. Our results are a factor of $1\times10^5$ ($3\times10^{5}$) more accurate for the E2 (E3) transition compared to previous measurements. The results open up the way to search for new physics via precise isotope shift measurements and improved tests of local Lorentz invariance using the metastable $^2F_{7/2}$ state of Yb$^+$., Comment: 6 pages, 5 figures

Sensory processing is often characterized as implementing probabilistic inference: networks of neurons compute posterior beliefs over unobserved causes given the sensory inputs. How these beliefs are computed and represented by neural responses is much-debated (Fiser et al. 2010, Pouget et al. 2013). A central debate concerns the question of whether neural responses represent samples of latent variables (Hoyer & Hyvarinnen 2003) or parameters of their distributions (Ma et al. 2006) with efforts being made to distinguish between them (Grabska-Barwinska et al. 2013). A separate debate addresses the question of whether neural responses are proportionally related to the encoded probabilities (Barlow 1969), or proportional to the logarithm of those probabilities (Jazayeri & Movshon 2006, Ma et al. 2006, Beck et al. 2012). Here, we show that these alternatives - contrary to common assumptions - are not mutually exclusive and that the very same system can be compatible with all of them. As a central analytical result, we show that modeling neural responses in area V1 as samples from a posterior distribution over latents in a linear Gaussian model of the image implies that those neural responses form a linear Probabilistic Population Code (PPC, Ma et al. 2006). In particular, the posterior distribution over some experimenter-defined variable like "orientation" is part of the exponential family with sufficient statistics that are linear in the neural sampling-based firing rates., Comment: First three contributed equally to the work

Questioning the presumably most basic assumptions about the structure of space and time has revolutionized our understanding of Nature. State-of-the-art atomic clocks make it possible to precisely test fundamental symmetry properties of spacetime, and search for physics beyond the standard model at low energy scales of just a few electron volts. Here, we experimentally demonstrate for the first time agreement of two single-ion clocks at the $10^{-18}$ level and directly confirm the validity of their uncertainty budgets over a half-year long comparison period. The two clock ions are confined in separate ion traps with quantization axes aligned along nonparallel directions. Hypothetical Lorentz symmetry violations would lead to sidereal modulations of the frequency offset. From the absence of such modulations at the $10^{-19}$ level we deduce stringent limits on Lorentz symmetry violation parameters for electrons in the range of $10^{-21}$, improving previous limits by two orders of magnitude., Comment: 21 pages, 2 figures, 2 tables

We devise a laser-controlled adaptive optical element which operates intracavity under high intensity radiation. This element substitutes a conventional mechanically deformable mirror and is free of critical heat-sensitive components and electronics. The deformation mechanism is based on the projection of a CW control laser onto a specially designed mirror. Mounted to a water-cooled heat sink, the mirror can handle laser radiation beyond 3 MW/cm^2. The properties of the adaptive optical element including the maximum correctable wavefront pitch of 800 nm are discussed. The successful implementation in a multipass thin disk amplifier is presented. An improvement of the beam quality by a factor of three is achieved. We identify measures to enhance the performance of the adaptive optic towards efficient operation in a high-power laser system.

We devise a perturbation-immune version of Ramsey's method of separated oscillatory fields. Spectroscopy of an atomic clock transition without compromising the clock's accuracy is accomplished by actively balancing the spectroscopic responses from phase-congruent Ramsey probe cycles of unequal durations. Our simple and universal approach eliminates a wide variety of interrogation-induced line shifts often encountered in high precision spectroscopy, among them, in particular, light shifts, phase chirps, and transient Zeeman shifts. We experimentally demonstrate auto-balanced Ramsey spectroscopy on the light shift prone $^{171}$Yb$^{+}$ electric octupole optical clock transition and show that interrogation defects are not turned into clock errors. This opens up frequency accuracy perspectives below the $10^{-18}$ level for the Yb$^{+}$ system and for other types of optical clocks., Comment: 8 pages, 6 figures

This brief bulletin addresses seven "myths" of tracking or ability grouping with gifted students in response to what is seen as a current movement against both ability grouping and tracking, which are characterized by the "anti-tracking" movement as inflexible and unaccommodating of a varied schedule. The following myths are countered: (1) tracking and ability grouping are the same thing; (2) ability grouping is elitist; (3) ability unfairly discriminates against racial and ethnic minority students because they are not equitably represented in gifted programs; (4) gifted students will make it on their own and grouping them by ability does not result in improved learning or achievement; (5) providing heterogeneously grouped cooperative learning experiences is most effective for serving all students, including the gifted; (6) assuring that there are some gifted students in all classrooms will provide positive role models for others and will automatically improve the classroom climate; and (7) all children at a certain grade level should be reading grade level material in a whole language learning environment. (Contains 25 references.) (DB)