Search

Establishing a unified theory of cognition has been a major goal of psychology. While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in its entirety. Here we introduce Centaur, a computational model that can predict and simulate human behavior in any experiment expressible in natural language. We derived Centaur by finetuning a state-of-the-art language model on a novel, large-scale data set called Psych-101. Psych-101 reaches an unprecedented scale, covering trial-by-trial data from over 60,000 participants performing over 10,000,000 choices in 160 experiments. Centaur not only captures the behavior of held-out participants better than existing cognitive models, but also generalizes to new cover stories, structural task modifications, and entirely new domains. Furthermore, we find that the model's internal representations become more aligned with human neural activity after finetuning. Taken together, Centaur is the first real candidate for a unified model of human cognition. We anticipate that it will have a disruptive impact on the cognitive sciences, challenging the existing paradigm for developing computational models.

The W. M. Keck Observatory Archive (KOA) has released the Observers Data Access Portal (ODAP), a web-application that delivers astronomical data from the W. M. Keck Observatory to the scheduled program's principal investigator and their collaborators anywhere in the world in near real-time. Data files and their associated metadata are streamed to a user's desktop machine moments after they are written to disk and archived in KOA. The ODAP User Interface is built in React and uses the WebSocket protocol to stream data between KOA and the user. This document describes the design of the tool, challenges encountered, shows how ODAP is integrated into the Keck observing model, and provides an analysis of usage metrics., Comment: 7 pages, 4 figures, SPIE Astronomical Telescopes + Instrumentation 2024

This paper describes the first Vertical Electron Cyclotron Emission (V-ECE) measurement of non-thermal electron distributions in the \textit{Tokamak \`a Configuration Variable}, TCV. These measurements were conducted in runaway electron scenarios and in the presence of Electron Cyclotron Current Drive. Measured intensities of linearly polarised X- and O-mode radiation from fast electrons allow the analysis of the energy distribution. The measurements were made possible through the creation of an operational regime for the diagnostic that is free of thermal background radiation, in relaxed electron density operations. This operational regime notably enables the cross-calibration of the diagnostic system, relying on thermal plasma measurements and modeling with the ray-tracing code SPECE., Comment: Submitted to: Plasma Phys. Control. Fusion

Because negative triangularity plasma scenarios remain in L-mode, they do not require external heating systems that exceed the H-mode power threshold. Operating with less heating has the potential to improve performance as heating generally degrades confinement in tokamaks. Using simple zero dimensional power balance and standard empirical scaling laws for confinement, we analyze the impact of external heating on several different reactor-relevant devices (i.e. SPARC, MANTA, ITER and DEMO). We compare the nominal externally heated scenarios against equivalent negative tringularity cases without external heating. For most of these devices, the Ohmically heated negative triangularity versions achieve better performance, particularly for devices with high magnetic field and/or high fusion gain.

In this paper, we take an axiomatic approach to defining a population hierarchical clustering for piecewise constant densities, and in a similar manner to Lebesgue integration, extend this definition to more general densities. When the density satisfies some mild conditions, e.g., when it has connected support, is continuous, and vanishes only at infinity, or when the connected components of the density satisfy these conditions, our axiomatic definition results in Hartigan's definition of cluster tree.

With active time-domain surveys like the Zwicky Transient Facility, the anticipated Rubin Observatory's Legacy Survey of Space and Time, and multi-messenger experiments such as LIGO/VIRGO/KANGRA for gravitational wave detection and IceCube for high-energy neutrino events, there is a new era in both time-domain and multi-messenger astronomy. The Astro2020 decadal survey highlights effectively responding to these astronomical alerts in a timely manner as a priority, and thus, there is an urgent need for the development of a seamless follow-up infrastructure at existing facilities that are capable of following up on detections at the survey depths. At the W. M. Keck Observatory (WMKO), we are actively constructing critical infrastructure, aimed at facilitating the Target-of-Opportunity (ToO) trigger, optimizing observational planning, streamlining data acquisition, and enhancing data product accessibility. In this document, we provide an overview of these developing services and place them in context of existing observatory infrastructure like the Keck Observatory Archive (KOA) and Data Services Initiative (DSI)., Comment: 9 pages, 3 figures, accompanying presentation at SPIE Astronomical Telescopes + Instrumentation 2024

Runaway electrons (REs) are a concern for tokamak fusion reactors from discharge startup to termination. A sudden localized loss of a multi-megaampere RE beam can inflict severe damage to the first wall. Should a disruption occur, the existence of a RE seed may play a significant role in the formation of a RE beam and the magnitude of its current. The application of central electron cyclotron resonance heating (ECRH) in the Tokamak \`a Configuration Variable (TCV) reduces an existing RE seed population by up to three orders of magnitude within only a few hundred milliseconds. Applying ECRH before a disruption can also prevent the formation of a post-disruption RE beam in TCV where it would otherwise be expected. The RE expulsion rate and consequent RE current reduction are found to increase with applied ECRH power. Whereas central ECRH is effective in expelling REs, off-axis ECRH has a comparatively limited effect. A simple 0-D model for the evolution of the RE population is presented that explains the effective ECRH-induced RE expulsion results from the combined effects of increased electron temperature and enhanced RE transport.

In this work, we study the impact of aspect ratio A = R0 /r (the ratio of major radius R0 to minor radius r) on the confinement benefits of Negative Triangularity (NT) plasma shaping. We use high-fidelity flux tube gyrokinetic GENE simulations and consider several different scenarios: four of them inspired by TCV experimental data, a scenario inspired by DIII-D experimental data and a scenario expected in the new SMART spherical tokamak. The present study reveals a surprising and non-trivial dependence. NT improves confinement at any value of A for ITG turbulence, while for TEM turbulence confinement is improved only in the case of large and conventional aspect ratios. Additionally, through a detailed study of a large aspect ratio case with pure ITG drive, we develop an intuitive physical picture that explains the beneficial effect of NT at large and conventional aspect ratios. This picture does not hold in TEM-dominated regimes, where a complex synergistic effect of many factors is found. Finally, we performed the first linear gyrokinetic simulations of SMART, finding that both NT and PT scenarios are dominated by micro-tearing-mode (MTM) turbulence and that NT is more susceptible to MTMs at tight aspect ratio. However, we found that a regime where ITG dominates in SMART can be found, and in this regime NT is more linearly stable.

Large language models (LLMs) have significantly advanced the field of artificial intelligence. Yet, evaluating them comprehensively remains challenging. We argue that this is partly due to the predominant focus on performance metrics in most benchmarks. This paper introduces CogBench, a benchmark that includes ten behavioral metrics derived from seven cognitive psychology experiments. This novel approach offers a toolkit for phenotyping LLMs' behavior. We apply CogBench to 35 LLMs, yielding a rich and diverse dataset. We analyze this data using statistical multilevel modeling techniques, accounting for the nested dependencies among fine-tuned versions of specific LLMs. Our study highlights the crucial role of model size and reinforcement learning from human feedback (RLHF) in improving performance and aligning with human behavior. Interestingly, we find that open-source models are less risk-prone than proprietary models and that fine-tuning on code does not necessarily enhance LLMs' behavior. Finally, we explore the effects of prompt-engineering techniques. We discover that chain-of-thought prompting improves probabilistic reasoning, while take-a-step-back prompting fosters model-based behaviors.

For all active instruments, the Keck Observatory Archive (KOA) now ingests raw data from the Keck Telescopes within 1 minute of acquisition, quick-look reduced data within 5 minutes of creation, and science ready reduced data for four instruments as they are created by their automated pipelines. On August 1, 2023, KOA released the Observers Data Access Portal (ODAP), which enables observers at the telescope and their collaborators anywhere in the world to securely monitor and download science, calibration, and quick-look data as they are ingested into the archive. The portal is built using Python Socket IO.WebSockets that ensure metadata appear in the portal as the data themselves are ingested. The portal itself is a dynamic web interface built with React. It enables users to view and customize metadata fields, filter metadata according to data type, and download data as they are ingested or in bulk through wget scripts. Observers have used the ODAP since its release and have provided feedback that will guide future releases., Comment: 4 pages, 2 figs, ADASS 2024 proceedings

Ecological rationality refers to the notion that humans are rational agents adapted to their environment. However, testing this theory remains challenging due to two reasons: the difficulty in defining what tasks are ecologically valid and building rational models for these tasks. In this work, we demonstrate that large language models can generate cognitive tasks, specifically category learning tasks, that match the statistics of real-world tasks, thereby addressing the first challenge. We tackle the second challenge by deriving rational agents adapted to these tasks using the framework of meta-learning, leading to a class of models called ecologically rational meta-learned inference (ERMI). ERMI quantitatively explains human data better than seven other cognitive models in two different experiments. It additionally matches human behavior on a qualitative level: (1) it finds the same tasks difficult that humans find difficult, (2) it becomes more reliant on an exemplar-based strategy for assigning categories with learning, and (3) it generalizes to unseen stimuli in a human-like way. Furthermore, we show that ERMI's ecologically valid priors allow it to achieve state-of-the-art performance on the OpenML-CC18 classification benchmark., Comment: 27 pages (9 pages of main text, 4 pages of references, and 14 pages of appendix), 13 figures, and 7 Tables

Characteristic classes, which are abstract topological invariants associated with vector bundles, have become an important notion in modern physics with surprising real-world consequences. As a representative example, the incredible properties of topological insulators, which are insulators in their bulk but conductors on their surface, can be completely characterized by a specific characteristic class associated with their electronic band structure, the first Chern class. Given their importance to next generation computing and the computational challenge of calculating them using first-principles approaches, there is a need to develop machine learning approaches to predict the characteristic classes associated with a material system. To aid in this program we introduce the {\emph{Haldane bundle dataset}}, which consists of synthetically generated complex line bundles on the $2$-torus. We envision this dataset, which is not as challenging as noisy and sparsely measured real-world datasets but (as we show) still difficult for off-the-shelf architectures, to be a testing ground for architectures that incorporate the rich topological and geometric priors underlying characteristic classes.

In recent years, negative triangularity (NT) has emerged as a potential high-confinement L-mode reactor solution. In this work, detachment is investigated using core density ramps in lower single null Ohmic L-mode plasmas across a wide range of upper, lower, and average triangularity (the mean of upper and lower triangularity: $\delta$) in the TCV tokamak. It is universally found that detachment is more difficult to access for NT shaping. The outer divertor leg of discharges with $\delta\approx -0.3$ could not be cooled to below $5~\mathrm{eV}$ through core density ramps alone. The behavior of the upstream plasma and geometrical divertor effects (e.g. a reduced connection length with negative lower triangularity) do not fully explain the challenges in detaching NT plasmas. Langmuir probe measurements of the target heat flux widths ($\lambda_q$) were constant to within 30% across an upper triangularity scan, while the spreading factor $S$ was lower by up to 50% for NT, indicating a generally lower integral Scrape-Off Layer width, $\lambda_{int}$. The line-averaged core density was typically higher for NT discharges for a given fuelling rate, possibly linked to higher particle confinement in NT. Conversely, the divertor neutral pressure and integrated particle fluxes to the targets were typically lower for the same line-averaged density, indicating that NT configurations may be closer to the sheath-limited regime than their PT counterparts, which may explain why NT is more challenging to detach.

Large Language Models (LLMs) are transforming society and permeating into diverse applications. As a result, LLMs will frequently interact with us and other agents. It is, therefore, of great societal value to understand how LLMs behave in interactive social settings. Here, we propose to use behavioral game theory to study LLM's cooperation and coordination behavior. To do so, we let different LLMs (GPT-3, GPT-3.5, and GPT-4) play finitely repeated games with each other and with other, human-like strategies. Our results show that LLMs generally perform well in such tasks and also uncover persistent behavioral signatures. In a large set of two players-two strategies games, we find that LLMs are particularly good at games where valuing their own self-interest pays off, like the iterated Prisoner's Dilemma family. However, they behave sub-optimally in games that require coordination. We, therefore, further focus on two games from these distinct families. In the canonical iterated Prisoner's Dilemma, we find that GPT-4 acts particularly unforgivingly, always defecting after another agent has defected only once. In the Battle of the Sexes, we find that GPT-4 cannot match the behavior of the simple convention to alternate between options. We verify that these behavioral signatures are stable across robustness checks. Finally, we show how GPT-4's behavior can be modified by providing further information about the other player as well as by asking it to predict the other player's actions before making a choice. These results enrich our understanding of LLM's social behavior and pave the way for a behavioral game theory for machines.

Large language models have shown tremendous performance in a variety of tasks. In-context learning -- the ability to improve at a task after being provided with a number of demonstrations -- is seen as one of the main contributors to their success. In the present paper, we demonstrate that the in-context learning abilities of large language models can be recursively improved via in-context learning itself. We coin this phenomenon meta-in-context learning. Looking at two idealized domains, a one-dimensional regression task and a two-armed bandit task, we show that meta-in-context learning adaptively reshapes a large language model's priors over expected tasks. Furthermore, we find that meta-in-context learning modifies the in-context learning strategies of such models. Finally, we extend our approach to a benchmark of real-world regression problems where we observe competitive performance to traditional learning algorithms. Taken together, our work improves our understanding of in-context learning and paves the way toward adapting large language models to the environment they are applied purely through meta-in-context learning rather than traditional finetuning.

Large language models (LLMs) are transforming research on machine learning while galvanizing public debates. Understanding not only when these models work well and succeed but also why they fail and misbehave is of great societal relevance. We propose to turn the lens of psychiatry, a framework used to describe and modify maladaptive behavior, to the outputs produced by these models. We focus on twelve established LLMs and subject them to a questionnaire commonly used in psychiatry. Our results show that six of the latest LLMs respond robustly to the anxiety questionnaire, producing comparable anxiety scores to humans. Moreover, the LLMs' responses can be predictably changed by using anxiety-inducing prompts. Anxiety-induction not only influences LLMs' scores on an anxiety questionnaire but also influences their behavior in a previously-established benchmark measuring biases such as racism and ageism. Importantly, greater anxiety-inducing text leads to stronger increases in biases, suggesting that how anxiously a prompt is communicated to large language models has a strong influence on their behavior in applied settings. These results demonstrate the usefulness of methods taken from psychiatry for studying the capable algorithms to which we increasingly delegate authority and autonomy.

Sessile droplets of a ferroelectric nematic liquid crystalline material were exposed to surface electric fields produced by pyroelectric and photogalvanic (photovoltaic) effects in X-cut iron-doped lithium niobate crystals. The resulting dynamic processes were monitored by polarization optical (video)microscopy (POM). During heating/cooling cycles, at first, the droplets change their shape from spherical to extended ellipsoidal. Then they start to move rapidly along the surface electric field, i.e., along the crystal's polar axis (c-axis). During this motion, several droplets merge into running streams (tendrils) extending towards the edges of the top surface area. Finally, practically all liquid crystalline material is transported from the top surface to the side surfaces of the crystal. At stabilized temperature, laser illumination of the assembly causes dynamic processes that are localized to the illuminated area. Also, in this case, the LC droplets merge into several tendril-like formations that are preferentially oriented along the c-axis of the crystal. The pattern of tendrils fluctuates with time, but it persists as long as the illumination is present. In this case, the LC material is transported between the central and the edge region of the illuminated area., Comment: 16 pages, 5 figures. arXiv admin note: text overlap with arXiv:2208.02318

A wide variety of adversarial attacks have been proposed and explored using image and audio data. These attacks are notoriously easy to generate digitally when the attacker can directly manipulate the input to a model, but are much more difficult to implement in the real-world. In this paper we present a universal, time invariant attack for general time series data such that the attack has a frequency spectrum primarily composed of the frequencies present in the original data. The universality of the attack makes it fast and easy to implement as no computation is required to add it to an input, while time invariance is useful for real-world deployment. Additionally, the frequency constraint ensures the attack can withstand filtering. We demonstrate the effectiveness of the attack in two different domains, speech recognition and unintended radiated emission, and show that the attack is robust against common transform-and-compare defense pipelines.

Due to the Internet of Things (IoT) proliferation, Radio Frequency (RF) channels are increasingly congested with new kinds of devices, which carry unique and diverse communication needs. This poses complex challenges in modern digital communications, and calls for the development of technological innovations that (i) optimize capacity (bitrate) in limited bandwidth environments, (ii) integrate cooperatively with already-deployed RF protocols, and (iii) are adaptive to the ever-changing demands in modern digital communications. In this paper we present methods for applying deep neural networks for spectral filling. Given an RF channel transmitting digital messages with a pre-established modulation scheme, we automatically learn novel modulation schemes for sending extra information, in the form of additional messages, "around" the fixed-modulation signals (i.e., without interfering with them). In so doing, we effectively increase channel capacity without increasing bandwidth. We further demonstrate the ability to generate signals that closely resemble the original modulations, such that the presence of extra messages is undetectable to third-party listeners. We present three computational experiments demonstrating the efficacy of our methods, and conclude by discussing the implications of our results for modern RF applications.

While it is not generally reflected in the `nice' datasets used for benchmarking machine learning algorithms, the real-world is full of processes that would be best described as many-to-many. That is, a single input can potentially yield many different outputs (whether due to noise, imperfect measurement, or intrinsic stochasticity in the process) and many different inputs can yield the same output (that is, the map is not injective). For example, imagine a sentiment analysis task where, due to linguistic ambiguity, a single statement can have a range of different sentiment interpretations while at the same time many distinct statements can represent the same sentiment. When modeling such a multivalued function $f: X \rightarrow Y$, it is frequently useful to be able to model the distribution on $f(x)$ for specific input $x$ as well as the distribution on fiber $f^{-1}(y)$ for specific output $y$. Such an analysis helps the user (i) better understand the variance intrinsic to the process they are studying and (ii) understand the range of specific input $x$ that can be used to achieve output $y$. Following existing work which used a fiber bundle framework to better model many-to-one processes, we describe how morphisms of fiber bundles provide a template for building models which naturally capture the structure of many-to-many processes.

Advanced manufacturing techniques have enabled the production of materials with state-of-the-art properties. In many cases however, the development of physics-based models of these techniques lags behind their use in the lab. This means that designing and running experiments proceeds largely via trial and error. This is sub-optimal since experiments are cost-, time-, and labor-intensive. In this work we propose a machine learning framework, differential property classification (DPC), which enables an experimenter to leverage machine learning's unparalleled pattern matching capability to pursue data-driven experimental design. DPC takes two possible experiment parameter sets and outputs a prediction of which will produce a material with a more desirable property specified by the operator. We demonstrate the success of DPC on AA7075 tube manufacturing process and mechanical property data using shear assisted processing and extrusion (ShAPE), a solid phase processing technology. We show that by focusing on the experimenter's need to choose between multiple candidate experimental parameters, we can reframe the challenging regression task of predicting material properties from processing parameters, into a classification task on which machine learning models can achieve good performance.

Results from the last FTU campaigns on the deuterium large (wrt FTU volume) pellet REs suppression capability, mainly due to the induced burst MHD activity expelling REs seed are presented for discharges with 0.5 MA and 5.3T. Clear indications of avalanche multiplication of REs following single pellet injection on 0.36 MA flat-top discharges is shown together with quantitative indications of dissipative effects in terms of critical electrical field increase due to fan-like instabilities. Analysis of large fan-like instabilities on post-disruption RE beams, that seem to be correlated with low electrical field and background density drops, reveal their strong RE energy suppression capability suggesting a new strategy for RE energy suppression controlling large fan instabilities. We demonstrate how such density drops can be induced using modulated ECRH power on post-disruption beams., Comment: DRAFT that will be submitted to NF (the conference version is on FEC/IAEA conference 2020)

We propose a polarization-independent optical isolator that does not use walk-off between the two orthogonal polarizations. The design is based on two Faraday rotators in combination with two half-wave plates in a closed, Sagnac-interferometer-like, configuration. An experimental prototype is tested successfully under variation of the input light polarization state with isolation level between 43 dB and 50 dB for all input polarizations (linear, circular, or elliptical).