Search

Much of today's machine learning research involves interpreting, modifying or visualizing models after they are trained. I present Penzai, a neural network library designed to simplify model manipulation by representing models as simple data structures, and Treescope, an interactive pretty-printer and array visualizer that can visualize both model inputs/outputs and the models themselves. Penzai models are built using declarative combinators that expose the model forward pass in the structure of the model object itself, and use named axes to ensure each operation is semantically meaningful. With Penzai's tree-editing selector system, users can both insert and replace model components, allowing them to intervene on intermediate values or make other edits to the model structure. Users can then get immediate feedback by visualizing the modified model with Treescope. I describe the motivation and main features of Penzai and Treescope, and discuss how treating the model as data enables a variety of analyses and interventions to be implemented as data-structure transformations, without requiring model designers to add explicit hooks., Comment: Presented at the ICML 2024 Mechanistic Interpretability workshop (Spotlight). 5 pages

Scattering-type scanning near-field optical microscopy (s-SNOM) is a powerful technique for extreme subwavelength imaging and spectroscopy, with around 20 nm spatial resolution. But quantitative relationships between experiment and material properties requires modelling, which can be computationally and conceptually challenging. In this work, we present snompy an open-source Python library which contains implementations of two of the most common s-SNOM models, the finite dipole model (FDM) and the point dipole model (PDM). We show a series of typical uses for this package with demonstrations including simulating nano-Fourier transform infrared (FTIR) spectra and recovering permittivity from experimental s-SNOM data. We also discuss the challenges faced with this sort of modelling, such as competing descriptions of the models in literature, and finite size effects. We hope that snompy will make quantitative s-SNOM modelling more accessible to the wider research community, which will further empower the use of s-SNOM for investigating nanoscale material properties., Comment: 26 pages, 8 figures

The CODEX-$\beta$ apparatus is a demonstrator for the proposed future CODEX-b experiment, a long-lived-particle detector foreseen for operation at IP8 during HL-LHC data-taking. The demonstrator project, intended to collect data in 2025, is described, with a particular focus on the design, construction, and installation of the new apparatus.

Strong gravitational lensing is a competitive tool to probe the dark matter and energy content of the universe. However, significant uncertainties can arise from the choice of lens model, and in particular the parameterisation of the line of sight. In this work, we consider the consequences of ignoring the contribution of foreground perturbers in lens modelling. We derive the explicit form of the degeneracy between the foreground shear and the ellipticity of a power law lens, which renders both quantities effectively unmeasurable from strong lensing observables. Nonetheless, we demonstrate that this degeneracy does not affect measurements of the Einstein radius. Foreground tidal effects are also not expected to bias the slope of the potential, and any biases in this slope should not affect the recovery of the Hubble constant. The foreground convergence term adds an additional uncertainty to the measurement of $H_0$, and we show that this uncertainty will be on the order of $1\%$ for lensing systems located along random lines of sight. There is evidence to indicate that the probability of strong lensing is higher towards overdense lines of sight, and this could result in a small systematic bias towards overestimations of $H_0$., Comment: 13 + 8 pages, 2 figures

It has been five years since the data sample from the LHCb detector, the first experiment optimized for heavy-flavor physics studies at a hadronic collider, was completed. These data led to many major discoveries in exotic hadron spectroscopy, which we review in this article. We supplement the experimental results with a selection of phenomenological interpretations. As the upgraded LHCb detector is expected to collect a larger data sample starting in 2024, the near- and further-future potential of the LHCb program in exotic hadron physics is also discussed.

This paper reports results from a high-fidelity driving simulator study (N=215) about a head-up display (HUD) that conveys a conditional automated vehicle's dynamic "uncertainty" about the current situation while fallback drivers watch entertaining videos. We compared (between-group) three design interventions: display (a bar visualisation of uncertainty close to the video), interruption (interrupting the video during uncertain situations), and combination (a combination of both), against a baseline (video-only). We visualised eye-tracking data to conduct a heatmap analysis of the four groups' gaze behaviour over time. We found interruptions initiated a phase during which participants interleaved their attention between monitoring and entertainment. This improved monitoring behaviour was more pronounced in combination compared to interruption, suggesting pre-warning interruptions have positive effects. The same addition had negative effects without interruptions (comparing baseline & display). Intermittent interruptions may have safety benefits over placing additional peripheral displays without compromising usability., Comment: Accepted for publication at the 2024 ACM Conference on Human Factors in Computing Systems (CHI'24)

Identifying how much a model ${\widehat{p}}_{\theta}(Y|X)$ knows about the stochastic real-world process $p(Y|X)$ it was trained on is important to ensure it avoids producing incorrect or "hallucinated" answers or taking unsafe actions. But this is difficult for generative models because probabilistic predictions do not distinguish between per-response noise (aleatoric uncertainty) and lack of knowledge about the process (epistemic uncertainty), and existing epistemic uncertainty quantification techniques tend to be overconfident when the model underfits. We propose a general strategy for teaching a model to both approximate $p(Y|X)$ and also estimate the remaining gaps between ${\widehat{p}}_{\theta}(Y|X)$ and $p(Y|X)$: train it to predict pairs of independent responses drawn from the true conditional distribution, allow it to "cheat" by observing one response while predicting the other, then measure how much it cheats. Remarkably, we prove that being good at cheating (i.e. cheating whenever it improves your prediction) is equivalent to being second-order calibrated, a principled extension of ordinary calibration that allows us to construct provably-correct frequentist confidence intervals for $p(Y|X)$ and detect incorrect responses with high probability. We demonstrate empirically that our approach accurately estimates how much models don't know across ambiguous image classification, (synthetic) language modeling, and partially-observable navigation tasks, outperforming existing techniques., Comment: Accepted at ICML 2024. 9 pages, 6 figures

Soft robotics is a rapidly growing field that spans the fields of chemistry, materials science, and engineering. Due to the diverse background of the field, there have been contrasting naming schemes such as 'intelligent', 'smart' and 'adaptive' materials which add vagueness to the broad innovation among literature. Therefore, a clear, functional and descriptive naming scheme is proposed in which a previously vague name -- Soft Material for Soft Actuators -- can remain clear and concise -- Phase-Change Elastomers for Artificial Muscles. By synthesizing the working principle, material, and application into a naming scheme, the searchability of soft robotics can be enhanced and applied to other fields. The field of thermo-active soft actuators spans multiple domains and requires added clarity. Thermo-active actuators have potential for a variety of applications spanning virtual reality haptics to assistive devices. This review offers a comprehensive guide to selecting the type of thermo-active actuator when one has an application in mind. Additionally, it discusses future directions and improvements that are necessary for implementation., Comment: 16 pages, 10 figures, accepted to Robotics Reports

Learning from human feedback (LHF) -- and in particular learning from pairwise preferences -- has recently become a crucial ingredient in training large language models (LLMs), and has been the subject of much research. Most recent works frame it as a reinforcement learning problem, where a reward function is learned from pairwise preference data and the LLM is treated as a policy which is adapted to maximize the rewards, often under additional regularization constraints. We propose an alternative interpretation which centers on the generative process for pairwise preferences and treats LHF as a density estimation problem. We provide theoretical and empirical results showing that for a family of generative processes defined via preference behavior distribution equations, training a reward function on pairwise preferences effectively models an annotator's implicit preference distribution. Finally, we discuss and present findings on "annotator misspecification" -- failure cases where wrong modeling assumptions are made about annotator behavior, resulting in poorly-adapted models -- suggesting that approaches that learn from pairwise human preferences could have trouble learning from a population of annotators with diverse viewpoints.

There are many physical processes that have inherent discontinuities in their mathematical formulations. This paper is motivated by the specific case of collisions between two rigid or deformable bodies and the intrinsic nature of that discontinuity. The impulse response to a collision is discontinuous with the lack of any response when no collision occurs, which causes difficulties for numerical approaches that require differentiability which are typical in machine learning, inverse problems, and control. We theoretically and numerically demonstrate that the derivative of the collision time with respect to the parameters becomes infinite as one approaches the barrier separating colliding from not colliding, and use lifting to complexify the solution space so that solutions on the other side of the barrier are directly attainable as precise values. Subsequently, we mollify the barrier posed by the unbounded derivatives, so that one can tunnel back and forth in a smooth and reliable fashion facilitating the use of standard numerical approaches. Moreover, we illustrate that standard approaches fail in numerous ways mostly due to a lack of understanding of the mathematical nature of the problem (e.g. typical backpropagation utilizes many rules of differentiation, but ignores L'Hopital's rule).

Various software efforts embrace the idea that object oriented programming enables a convenient implementation of the chain rule, facilitating so-called automatic differentiation via backpropagation. Such frameworks have no mechanism for simplifying the expressions (obtained via the chain rule) before evaluating them. As we illustrate below, the resulting errors tend to be unbounded.

Many ML applications and products train on medium amounts of input data but get bottlenecked in real-time inference. When implementing ML systems, conventional wisdom favors segregating ML code into services queried by product code via Remote Procedure Call (RPC) APIs. This approach clarifies the overall software architecture and simplifies product code by abstracting away ML internals. However, the separation adds network latency and entails additional CPU overhead. Hence, we simplify inference algorithms and embed them into the product code to reduce network communication. For public datasets and a high-performance real-time platform that deals with tabular data, we show that over half of the inputs are often amenable to such optimization, while the remainder can be handled by the original model. By applying our optimization with AutoML to both training and inference, we reduce inference latency by 1.3x, CPU resources by 30%, and network communication between application front-end and ML back-end by about 50% for a commercial end-to-end ML platform that serves millions of real-time decisions per second.

Large language models show impressive results at predicting structured text such as code, but also commonly introduce errors and hallucinations in their output. When used to assist software developers, these models may make mistakes that users must go back and fix, or worse, introduce subtle bugs that users may miss entirely. We propose Randomized Utility-driven Synthesis of Uncertain REgions (R-U-SURE), an approach for building uncertainty-aware suggestions based on a decision-theoretic model of goal-conditioned utility, using random samples from a generative model as a proxy for the unobserved possible intents of the end user. Our technique combines minimum-Bayes-risk decoding, dual decomposition, and decision diagrams in order to efficiently produce structured uncertainty summaries, given only sample access to an arbitrary generative model of code and an optional AST parser. We demonstrate R-U-SURE on three developer-assistance tasks, and show that it can be applied different user interaction patterns without retraining the model and leads to more accurate uncertainty estimates than token-probability baselines. We also release our implementation as an open-source library at https://github.com/google-research/r_u_sure., Comment: To appear at ICML 2023. 9 pages, 6 figures

Contrastive learning is a powerful framework for learning self-supervised representations that generalize well to downstream supervised tasks. We show that multiple existing contrastive learning methods can be reinterpreted as learning kernel functions that approximate a fixed positive-pair kernel. We then prove that a simple representation obtained by combining this kernel with PCA provably minimizes the worst-case approximation error of linear predictors, under a straightforward assumption that positive pairs have similar labels. Our analysis is based on a decomposition of the target function in terms of the eigenfunctions of a positive-pair Markov chain, and a surprising equivalence between these eigenfunctions and the output of Kernel PCA. We give generalization bounds for downstream linear prediction using our Kernel PCA representation, and show empirically on a set of synthetic tasks that applying Kernel PCA to contrastive learning models can indeed approximately recover the Markov chain eigenfunctions, although the accuracy depends on the kernel parameterization as well as on the augmentation strength., Comment: Published at ICLR 2023

This is the Snowmass2021 Energy Frontier (EF) Beyond the Standard Model (BSM) report. It combines the EF topical group reports of EF08 (Model-specific explorations), EF09 (More general explorations), and EF10 (Dark Matter at Colliders). The report includes a general introduction to BSM motivations and the comparative prospects for proposed future experiments for a broad range of potential BSM models and signatures, including compositeness, SUSY, leptoquarks, more general new bosons and fermions, long-lived particles, dark matter, charged-lepton flavor violation, and anomaly detection., Comment: 108 pages + 38 pages references and appendix, 37 figures, Report of the Topical Group on Beyond the Standard Model Physics at Energy Frontier for Snowmass 2021. The first nine authors are the Conveners, with Contributions from the other authors

Graph representations of programs are commonly a central element of machine learning for code research. We introduce an open source Python library python_graphs that applies static analysis to construct graph representations of Python programs suitable for training machine learning models. Our library admits the construction of control-flow graphs, data-flow graphs, and composite ``program graphs'' that combine control-flow, data-flow, syntactic, and lexical information about a program. We present the capabilities and limitations of the library, perform a case study applying the library to millions of competitive programming submissions, and showcase the library's utility for machine learning research., Comment: 21 pages, 14 figures

In this Snowmass contribution we present a comprehensive status update on the progress and plans for the proposed CODEX-b detector, intended to search for long-lived particles beyond the Standard Model. We review the physics case for the proposal and present recent progress on optimization strategies for the detector and shielding design, as well as the development of new fast and full simulation frameworks. A summary of the technical design for a smaller demonstrator detector (CODEX-$\beta$) for the upcoming Run~3 of the LHC is also discussed, alongside the road towards realization of the full experiment at the High-Luminosity LHC., Comment: 13 pages, 13 figures, Snowmass contribution