Search

This paper deals with a new design methodology for stratified comparative experiments based on interacting reinforced urn systems. The key idea is to model the interaction between urns for borrowing information across strata and to use it in the design phase in order to i) enhance the information exchange at the beginning of the study, when only few subjects have been enrolled and the stratum-specific information on treatments' efficacy could be scarce, ii) let the information sharing adaptively evolves via a reinforcement mechanism based on the observed outcomes, for skewing at each step the allocations towards the stratum-specific most promising treatment and iii) make the contribution of the strata with different treatment efficacy vanishing as the stratum information grows. In particular, we introduce the Interacting Urns Design, namely a new Covariate-Adjusted Response-Adaptive procedure, that randomizes the treatment allocations according to the evolution of the urn system. The theoretical properties of this proposal are described and the corresponding asymptotic inference is provided. Moreover, by a functional central limit theorem, we obtain the asymptotic joint distribution of the Wald-type sequential test statistics, which allows to sequentially monitor the suggested design in the clinical practice.

Using high-throughput automation of ab-initio impurity embedding simulations, we created a database of $3d$ and $4d$ transition metal defects embedded into the prototypical topological insulator (TI) Bi$_2$Te$_3$. We simulate both single impurities as well as impurity dimers at different impurity-impurity distances inside the TI. We extract changes to magnetic moments, analyze the polarizability of non-magnetic impurity atoms via nearby magnetic impurity atoms and calculate the exchange coupling constants for a Heisenberg Hamiltonian. We uncover chemical trends in the exchange coupling constants and discuss the impurities' potential with respect to magnetic order in the fields of quantum anomalous Hall insulators and topological quantum computing. In particular, we predict that co-doping of different magnetic dopants is a viable strategy to engineer the magnetic ground state in magnetic TIs.

Low-rank adapters (LoRA) and their variants are popular parameter-efficient fine-tuning (PEFT) techniques that closely match full model fine-tune performance while requiring only a small number of additional parameters. These additional LoRA parameters are specific to the base model being adapted. When the base model needs to be deprecated and replaced with a new one, all the associated LoRA modules need to be re-trained. Such re-training requires access to the data used to train the LoRA for the original base model. This is especially problematic for commercial cloud applications where the LoRA modules and the base models are hosted by service providers who may not be allowed to host proprietary client task data. To address this challenge, we propose $\textit{Trans-LoRA}$ -- a novel method for lossless, nearly data-free transfer of LoRAs across base models. Our approach relies on synthetic data to transfer LoRA modules. Using large language models, we design a synthetic data generator to approximate the data-generating process of the $\textit{observed}$ task data subset. Training on the resulting synthetic dataset transfers LoRA modules to new models. We show the effectiveness of our approach using both LLama and Gemma model families. Our approach achieves lossless (mostly improved) LoRA transfer between models within and across different base model families, and even between different PEFT methods, on a wide variety of tasks.

On-device machine learning (ODML) enables powerful edge applications, but power consumption remains a key challenge for resource-constrained devices. To address this, developers often face a trade-off between model accuracy and power consumption, employing either computationally intensive models on high-power cores or pared-down models on low-power cores. Both approaches typically lead to a compromise in user experience (UX). This work focuses on the use of Gated Compression (GC) layer to enhance ODML model performance while conserving power and maximizing cost-efficiency, especially for always-on use cases. GC layers dynamically regulate data flow by selectively gating activations of neurons within the neural network and effectively filtering out non-essential inputs, which reduces power needs without compromising accuracy, and enables more efficient execution on heterogeneous compute cores. These improvements enhance UX through prolonged battery life, improved device responsiveness, and greater user comfort. In this work, we have integrated GC layers into vision and speech domain models including the transformer-based ViT model. Our experiments demonstrate theoretical power efficiency gains ranging from 158x to 30,000x for always-on scenarios. This substantial improvement empowers ODML applications with enhanced UX benefits., Comment: Initial Submission

This paper studies the relationship between the surface form of a mathematical problem and its solvability by large language models. We find that subtle alterations in the surface form can significantly impact the answer distribution and the solve rate, exposing the language model's lack of robustness and sensitivity to the surface form in reasoning through complex problems. To improve mathematical reasoning performance, we propose Self-Consistency-over-Paraphrases (SCoP), which diversifies reasoning paths from specific surface forms of the problem. We evaluate our approach on four mathematics reasoning benchmarks over three large language models and show that SCoP improves mathematical reasoning performance over vanilla self-consistency, particularly for problems initially deemed unsolvable. Finally, we provide additional experiments and discussion regarding problem difficulty and surface forms, including cross-model difficulty agreement and paraphrasing transferability, and Variance of Variations (VOV) for language model evaluation., Comment: Accepted to the main conference of NAACL (2024)

On-device machine learning (ODML) enables intelligent applications on resource-constrained devices. However, power consumption poses a major challenge, forcing a trade-off between model accuracy and power efficiency that often limits model complexity. The previously established Gated Compression (GC) layers offer a solution, enabling power efficiency without sacrificing model performance by selectively gating samples that lack signals of interest. However, their reliance on ground truth labels limits GC layers to supervised tasks. This work introduces the Dynamic Switch Layer (DSL), extending the benefits of GC layers to unsupervised learning scenarios, and maintaining power efficiency without the need for labeled data. The DSL builds upon the GC architecture, leveraging a dynamic pathway selection, and adapting model complexity in response to the innate structure of the data. We integrate the DSL into the SoundStream architecture and demonstrate that by routing up to 80% of samples through a lightweight pass we achieve a 12.3x reduction in the amount of computation performed and a 20.9x reduction in model size. This reduces the on-device inference latency by up to 26.5% and improves power efficiency by up to 21.4% without impacting model performance., Comment: Initial Submission

In applications of optical multipass cells in photochemical reactors and laser excitation of weak transitions, estimation of the radiation dose in a volume of interest allows to assess the performance and optimize the design of the cell. We adopt radiant fluence as the figure of merit and employ the radiative transfer equation to derive analytical expressions for average radiant fluence in a given volume of interest. These are used to establish practical approximations and a Monte Carlo ray tracing model of the spatial distribution of fluence. Ray tracing is performed with Zemax OpticsStudio 18.9., Comment: 10 pages, 5 figures

Ultra-high dose rate radiation therapy (FLASH) based on proton irradiation is of major interest for cancer treatments but creates new challenges for dose monitoring. Amorphous hydrogenated silicon is known to be one of the most radiation-hard semiconductors. In this study, detectors based on this material are investigated at proton dose rates similar to or exceeding those required for FLASH therapy. Tested detectors comprise two different types of contacts, two different thicknesses deposited either on glass or on polyimide substrates. All detectors exhibit excellent linear behaviour as a function of dose rate up to a value of 20 kGy/s. Linearity is achieved independently of the depletion condition of the device and remarkably in passive (unbiased) conditions. The degradation of the performance as a function of the dose rate and its recovery are also discussed., Comment: 16 pages, 9 figures, presented at 29th Internation Conference on Solid-State Dosimetry, 2023

We present Deep Search DocQA. This application enables information extraction from documents via a question-answering conversational assistant. The system integrates several technologies from different AI disciplines consisting of document conversion to machine-readable format (via computer vision), finding relevant data (via natural language processing), and formulating an eloquent response (via large language models). Users can explore over 10,000 Environmental, Social, and Governance (ESG) disclosure reports from over 2000 corporations. The Deep Search platform can be accessed at: https://ds4sd.github.io., Comment: Accepted at the Demonstration Track of the 38th Annual AAAI Conference on Artificial Intelligence (AAAI 24)

Hydrogenated amorphous silicon (a-Si:H) is a material having an intrinsically high radiation hardness that can be deposited on flexible substrates like Polyimide. For these properties a-Si:H can be used for the production of flexible sensors. a-Si:H sensors can be successfully utilized in dosimetry, beam monitoring for particle physics (x-ray, electron, gamma-ray and proton detection) and radiotherapy, radiation flux measurement for space applications (study of solar energetic particles and stellar events) and neutron flux measurements. In this paper we have studied the dosimetric x-ray response of n-i-p diodes deposited on Polyimide. We measured the linearity of the photocurrent response to x-rays versus dose-rate from which we have extracted the dosimetric x-ray sensitivity at various bias voltages. In particular low bias voltage operation has been studied to assess the high energy efficiency of these kind of sensor. A measurement of stability of x-ray response versus time has been shown. The effect of detectors annealing has been studied. Operation under bending at various bending radii is also shown.

We report on an Yb:YAG thin-disk multipass amplifier delivering 50 ns long pulses at a central wavelength of 1030 nm with an energy of 330 mJ at a repetition rate of 100 Hz. The beam quality factor at the maximum energy was measured to be $\text{M}^2 = 1.17$. The small signal gain is 20, and the gain at 330 mJ was measured to be 6.9. The 20-pass amplifier is designed as a concatenation of stable resonator segments in which the beam is alternately Fourier transformed and relay-imaged back to the disk by a 4f-imaging optical scheme stage. The Fourier transform propagation makes the output beam robust against spherical phase front distortions, while the 4f-stage is used to compensate the thermal lens of the thin-disk and to reduce the footprint of the amplifier.

Interfacing superconductors with magnetic or topological materials offers a playground where novel phenomena like topological superconductivity, Majorana zero modes, or superconducting spintronics are emerging. In this work, we discuss recent developments in the Kohn-Sham Bogoliubov-de Gennes method, which allows to perform material-specific simulations of complex superconducting heterostructures on the basis of density functional theory. As a model system we study magnetically-doped Pb. In our analysis we focus on the interplay of magnetism and superconductivity. This combination leads to Yu-Shiba-Rusinov (YSR) in-gap bound states at magnetic defects and the breakdown of superconductivity at larger impurity concentrations. Moreover, the influence of spin-orbit coupling and on orbital splitting of YSR states as well as the appearance of a triplet component in the order parameter is discussed. These effects can be exploited in S/F/S-type devices (S=superconductor, F=ferromagnet) in the field of superconducting spintronics.

We demonstrate an injection-seeded thin-disk Yb:YAG laser at 1030 nm, stabilized by the Pound-Drever-Hall (PDH) method. We modified the PDH scheme to obtain an error signal free from Trojan locking points, which allowed robust re-locking of the laser and reliable long-term operation. The single-frequency pulses have 50 mJ energy (limited to avoid laser-induced damage) with a beam quality of $\text{M}^2$ < 1.1 and an adjustable length of 55-110 ns. Heterodyne measurements confirmed a spectral linewidth of 3.7 MHz. The short pulse build-up time (850 ns) makes this laser suitable for laser spectroscopy of muonic hydrogen, pursued by the CREMA collaboration.

Extracting dense representations for terms and phrases is a task of great importance for knowledge discovery platforms targeting highly-technical fields. Dense representations are used as features for downstream components and have multiple applications ranging from ranking results in search to summarization. Common approaches to create dense representations include training domain-specific embeddings with self-supervised setups or using sentence encoder models trained over similarity tasks. In contrast to static embeddings, sentence encoders do not suffer from the out-of-vocabulary (OOV) problem, but impose significant computational costs. In this paper, we propose a fully unsupervised approach to text encoding that consists of training small character-based models with the objective of reconstructing large pre-trained embedding matrices. Models trained with this approach can not only match the quality of sentence encoders in technical domains, but are 5 times smaller and up to 10 times faster, even on high-end GPUs., Comment: Accepted at ACL 2023 (industry). 10 pages, 3 figures, 5 tables

Hydrogen-like light muonic ions, in which one negative muon replaces all the electrons, are extremely sensitive probes of nuclear structure, because the large muon mass increases tremendously the wave function overlap with the nucleus. Using pulsed laser spectroscopy we have measured three 2S-2P transitions in the muonic helium-3 ion ($\mu^3$He$^+$), an ion formed by a negative muon and bare helium-3 nucleus. This allowed us to extract the Lamb shift $E(2P_{1/2}-2S_{1/2})= 1258.598(48)^{\rm exp}(3)^{\rm theo}$ meV, the 2P fine structure splitting $E_{\rm FS}^{\rm exp} = 144.958(114)$ meV, and the 2S-hyperfine splitting (HFS) $E_{\rm HFS}^{\rm exp} = -166.495(104)^{\rm exp}(3)^{\rm theo}$ meV in $\mu^3$He$^+$. Comparing these measurements to theory we determine the rms charge radius of the helion ($^3$He nucleus) to be $r_h$ = 1.97007(94) fm. This radius represents a benchmark for few nucleon theories and opens the way for precision tests in $^3$He atoms and $^3$He-ions. This radius is in good agreement with the value from elastic electron scattering, but a factor 15 more accurate. Combining our Lamb shift measurement with our earlier one in $\mu^4$He$^+$ we obtain $r_h^2-r_\alpha^2 = 1.0636(6)^{\rm exp}(30)^{\rm theo}$ fm$^2$ to be compared to results from the isotope shift measurements in regular He atoms, which are however affected by long-standing tensions. By comparing $E_{\rm HFS}^{\rm exp}$ with theory we also obtain the two-photon-exchange contribution (including higher orders) which is another important benchmark for ab-initio few-nucleon theories aiming at understanding the magnetic and current structure of light nuclei., Comment: 8 pages, 5 figures

The CREMA collaboration is pursuing a measurement of the ground-state hyperfine splitting (HFS) in muonic hydrogen ($\mu$p) with 1 ppm accuracy by means of pulsed laser spectroscopy. In the proposed experiment, the $\mu$p atom is excited by a laser pulse from the singlet to the triplet hyperfine sub-levels, and is quenched back to the singlet state by an inelastic collision with a H$_2$ molecule. The resulting increase of kinetic energy after this cycle modifies the $\mu$p atom diffusion in the hydrogen gas and the arrival time of the $\mu$p atoms at the target walls. This laser-induced modification of the arrival times is used to expose the atomic transition. In this paper we present the simulation of the $\mu$p diffusion in the H$_2$ gas which is at the core of the experimental scheme. These simulations have been implemented with the Geant4 framework by introducing various low-energy processes including the motion of the H$_2$ molecules, i.e. the effects related with the hydrogen target temperature. The simulations have been used to optimize the hydrogen target parameters (pressure, temperatures and thickness) and to estimate signal and background rates. These rates allow to estimate the maximum time needed to find the resonance and the statistical accuracy of the spectroscopy experiment., Comment: Submission to SciPost

Recent progress in laser and x-ray spectroscopy of muonic atoms offers promising long-term possibilities at the intersection of atomic, nuclear and particle physics. In muonic hydrogen, laser spectroscopy measurements will determine the ground-state hyperfine splitting (HFS) and additionally improve the Lamb shift by a factor of 5. Precision spectroscopy with cryogenic microcalorimeters has the potential to significantly improve the charge radii of the light nuclei in the $Z=3-8$ range. Complementary progress in precision should be achieved on the theory of nucleon- and nuclear-structure effects. The impact of this muonic-atom spectroscopy program will be amplified by the upcoming results from H and He$^+$ spectroscopy, simple molecules such as HD$^+$ and Penning trap measurements. In this broader context, one can test ab-initio nuclear theories, bound-state QED for two- or three-body systems, and determine fundamental constants, such as the Rydberg ($R_\infty$) and the fine-structure ($\alpha$) constants., Comment: 8 pages, 2 figures, submitted as community input for the NuPECC Long Range Plan 2024 (https://indico.ph.tum.de/event/7050/)

Term extraction is an information extraction task at the root of knowledge discovery platforms. Developing term extractors that are able to generalize across very diverse and potentially highly technical domains is challenging, as annotations for domains requiring in-depth expertise are scarce and expensive to obtain. In this paper, we describe the term extraction subsystem of a commercial knowledge discovery platform that targets highly technical fields such as pharma, medical, and material science. To be able to generalize across domains, we introduce a fully unsupervised annotator (UA). It extracts terms by combining novel morphological signals from sub-word tokenization with term-to-topic and intra-term similarity metrics, computed using general-domain pre-trained sentence-encoders. The annotator is used to implement a weakly-supervised setup, where transformer-models are fine-tuned (or pre-trained) over the training data generated by running the UA over large unlabeled corpora. Our experiments demonstrate that our setup can improve the predictive performance while decreasing the inference latency on both CPUs and GPUs. Our annotators provide a very competitive baseline for all the cases where annotations are not available., Comment: Accepted at EMNLP 2022 (industry). 8 pages, 3 figures, 3 tables

Aging civil infrastructures are closely monitored by engineers for damage and critical defects. As the manual inspection of such large structures is costly and time-consuming, we are working towards fully automating the visual inspections to support the prioritization of maintenance activities. To that end we combine recent advances in drone technology and deep learning. Unfortunately, annotation costs are incredibly high as our proprietary civil engineering dataset must be annotated by highly trained engineers. Active learning is, therefore, a valuable tool to optimize the trade-off between model performance and annotation costs. Our use-case differs from the classical active learning setting as our dataset suffers from heavy class imbalance and consists of a much larger already labeled data pool than other active learning research. We present a novel method capable of operating in this challenging setting by replacing the traditional active learning acquisition function with an auxiliary binary discriminator. We experimentally show that our novel method outperforms the best-performing traditional active learning method (BALD) by 5% and 38% accuracy on CIFAR-10 and our proprietary dataset respectively.

The Pound-Drever-Hall (PDH) technique is a popular method for stabilizing the frequency of a laser to a stable optical resonator or, vice versa, the length of a resonator to the frequency of a stable laser. We propose a refinement of the technique yielding an "infinite" dynamic (capture) range so that a resonator is correctly locked to the seed frequency, even after large perturbations. The stable but off-resonant lock points (also called Trojan operating points), present in conventional PDH error signals, are removed by phase modulating the seed laser at a frequency corresponding to half the free spectral range of the resonator. We verify the robustness of our scheme experimentally by realizing an injection-seeded Yb:YAG thin-disk laser. We also give an analytical formulation of the PDH error signal for arbitrary modulation frequencies and discuss the parameter range for which our PDH locking scheme guarantees correct locking. Our scheme is simple as it does not require additional electronics apart from the standard PDH setup and is particularly suited to realize injection-seeded lasers and injection-seeded optical parametric oscillators., Comment: 8 pages, 11 figures

Muonic atom spectroscopy -- the measurement of the x rays emitted during the formation process of a muonic atom -- has a long standing history in probing the shape and size of nuclei. In fact, almost all stable elements have been subject to muonic atom spectroscopy measurements and the absolute charge radii extracted from these measurements typically offer the highest accuracy available. However, so far only targets of at least a few hundred milligram could be used as it required to stop a muon beam directly in the target to form the muonic atom. We have developed a new method relying on repeated transfer reactions taking place inside a 100-bar hydrogen gas cell with an admixture of 0.25% deuterium that allows us to drastically reduce the amount of target material needed while still offering an adequate efficiency. Detailed simulations of the transfer reactions match the measured data, suggesting good understanding of the processes taking place inside the gas mixture. As a proof of principle we demonstrate the method with a measurement of the 2p-1s muonic x rays from a 5-{\mu}g gold target., Comment: 16 pages, 14 figures

A low-emittance, high-intensity atomic beam of muonium ($\mathrm{M}=\mu^+ + \mathrm{e}^-$) using superfluid helium as muon-to-muonium converter is being developed at the Paul Scherrer Institute (PSI). This beam could advance laser spectroscopy of muonium and allow the first atomic interferometry experiments for the direct observation of the M gravitational interaction. In this paper, we describe the development of compact detection schemes which resulted in the background-suppressed observation of atomic muonium in vacuum, and can be adapted for cryogenic measurements. Using these setups, we compared the emission characteristics of various muonium production targets using low momentum ($p_{\mu} = 11$-$13~$MeV/c) muons, and observed muonium emission from zeolite targets into vacuum for the first time. For a specific laser-ablated aerogel target, we determined a muon-to-vacuum-muonium conversion efficiency of $7.23 \pm 0.05 \text{(stat)} ^{+1.06}_{-0.76}\text{(sys)}\,\text{%}$, assuming thermal emission of muonium. Moreover, we investigated muonium-helium collisions and from it we determined an upper temperature limit of 0.3 K for the superfluid helium converter.

We present a high-precision calculation of the recoil--finite-size correction to the hyperfine splitting (HFS) in muonic and electronic hydrogen based on nucleon electromagnetic form factors obtained from dispersion theory. This will help guide the upcoming searches of the HFS transition in muonic hydrogen, and will allow a precise determination of the polarizability and Zemach radius contributions when this transition is found., Comment: 6 pages, 2 figures

Laser spectroscopy of muonic atoms has been recently used to probe properties of light nuclei with unprecedented precision. We introduce nuclear effects in hydrogen-like atoms, nucleon structure quantities (form factors, structure functions, polarizabilities) and their effects in the Lamb shift and hyperfine splitting (HFS) of muonic hydrogen ($\mu$H). Updated theory predictions for the Lamb shift and HFS in $\mu$H are presented. We review the challenges of the ongoing effort to measure the ground-state HFS in $\mu$H and its impact on our understanding of the nucleon spin structure. To narrow down this search, we present a novel theory prediction obtained by scaling the measured HFS in hydrogen leveraging radiative corrections. We also summarize recent developments in the spectroscopy of simple atomic and molecular systems and emphasize how they allow for precise determinations of fundamental constants, bound-state QED tests and New Physics searches., Comment: 48 pages; 11 figures; 4 tables; final version with extended Supplement; posted with permission from the Annual Review of Nuclear and Particle Science, Volume 72 by Annual Reviews, http://www.annualreviews.org

Artificial intelligence and machine learning algorithms have become ubiquitous. Although they offer a wide range of benefits, their adoption in decision-critical fields is limited by their lack of interpretability, particularly with textual data. Moreover, with more data available than ever before, it has become increasingly important to explain automated predictions. Generally, users find it difficult to understand the underlying computational processes and interact with the models, especially when the models fail to generate the outcomes or explanations, or both, correctly. This problem highlights the growing need for users to better understand the models' inner workings and gain control over their actions. This dissertation focuses on two fundamental challenges of addressing this need. The first involves explanation generation: inferring high-quality explanations from text documents in a scalable and data-driven manner. The second challenge consists in making explanations actionable, and we refer to it as critiquing. This dissertation examines two important applications in natural language processing and recommendation tasks. Overall, we demonstrate that interpretability does not come at the cost of reduced performance in two consequential applications. Our framework is applicable to other fields as well. This dissertation presents an effective means of closing the gap between promise and practice in artificial intelligence., Comment: Ph.D. Thesis, Ecole Polytechnique F\'ed\'erale de Lausanne (EPFL). See https://infoscience.epfl.ch/record/292597 for the original version. This current version fixes two references

Explanation is important for text classification tasks. One prevalent type of explanation is rationales, which are text snippets of input text that suffice to yield the prediction and are meaningful to humans. A lot of research on rationalization has been based on the selective rationalization framework, which has recently been shown to be problematic due to the interlocking dynamics. In this paper, we show that we address the interlocking problem in the multi-aspect setting, where we aim to generate multiple rationales for multiple outputs. More specifically, we propose a multi-stage training method incorporating an additional self-supervised contrastive loss that helps to generate more semantically diverse rationales. Empirical results on the beer review dataset show that our method improves significantly the rationalization performance., Comment: Work in progress. 5 pages, 2 figures, 1 table

There has recently been growing interest in the automatic generation of cooking recipes that satisfy some form of dietary restrictions, thanks in part to the availability of online recipe data. Prior studies have used pre-trained language models, or relied on small paired recipe data (e.g., a recipe paired with a similar one that satisfies a dietary constraint). However, pre-trained language models generate inconsistent or incoherent recipes, and paired datasets are not available at scale. We address these deficiencies with RecipeCrit, a hierarchical denoising auto-encoder that edits recipes given ingredient-level critiques. The model is trained for recipe completion to learn semantic relationships within recipes. Our work's main innovation is our unsupervised critiquing module that allows users to edit recipes by interacting with the predicted ingredients; the system iteratively rewrites recipes to satisfy users' feedback. Experiments on the Recipe1M recipe dataset show that our model can more effectively edit recipes compared to strong language-modeling baselines, creating recipes that satisfy user constraints and are more correct, serendipitous, coherent, and relevant as measured by human judges., Comment: Accepted at EACL 2023. 10 pages, 2 figures, 6 tables, 1 algorithm

Providing explanations for recommended items allows users to refine the recommendations by critiquing parts of the explanations. As a result of revisiting critiquing from the perspective of multimodal generative models, recent work has proposed M&Ms-VAE, which achieves state-of-the-art performance in terms of recommendation, explanation, and critiquing. M&Ms-VAE and similar models allow users to negatively critique (i.e., explicitly disagree). However, they share a significant drawback: users cannot positively critique (i.e., highlight a desired feature). We address this deficiency with M&Ms-VAE+, an extension of M&Ms-VAE that enables positive and negative critiquing. In addition to modeling users' interactions and keyphrase-usage preferences, we model their keyphrase-usage dislikes. Moreover, we design a novel critiquing module that is trained in a self-supervised fashion. Our experiments on two datasets show that M&Ms-VAE+ matches or exceeds M&Ms-VAE in recommendation and explanation performance. Furthermore, our results demonstrate that representing positive and negative critiques differently enables M&Ms-VAE+ to significantly outperform M&Ms-VAE and other models in positive and negative multi-step critiquing., Comment: 5 pages, 2 figures, 2 tables

A new technique for detection of slow neutrons with gaseous detectors using ultra-thin layers with 10B atoms is presented. The reaction between a thermal neutron and a 10B atom releases 2 secondary particles, namely a 7Li ion and an alpha particle, which due to momentum conservation are emitted in opposite directions, along the same line (back to back). Current boron coated neutron detectors are equipped with 10B films with thicknesses of several micrometers, deposited on very thick substrate plates. However, since the ranges of the 7Li ion and the alpha particle are of few micrometeres in most materials, one of these particles is always lost in the 10B layer or substrate. As such, these detectors lose the ability to reconstruct the reaction line of action and to precisely determine the neutron position, as only one of the two secondary particles track can be measured. With the technique now presented, the sum of the 10B layer and the substrate thicknesses is small enough to allow for both secondary particles to escape and ionize the gas in opposite sides of the 10B converter foil. Independent readout structures, one on each side of the 10B converter foil, detect each secondary particle and determine its track centroid and the deposited energy. Since the two secondary particles are emitted back to back, the neutron position can be obtained by combining the information recorded by the two readout structures. Through GEANT4 simulations, we verified that the spatial resolution can be significantly improved: our results show that, by using a B4C layer with a thickness of 1 micron on a 0.9 micron Mylar substrate, the spatial resolution can by improved by a factor of 8, compared to conventional detectors with thick 10B detection layers.

Large pre-trained language models based on transformer architecture have drastically changed the natural language processing (NLP) landscape. However, deploying those models for on-device applications in constrained devices such as smart watches is completely impractical due to their size and inference cost. As an alternative to transformer-based architectures, recent work on efficient NLP has shown that weight-efficient models can attain competitive performance for simple tasks, such as slot filling and intent classification, with model sizes in the order of the megabyte. This work introduces the pNLP-Mixer architecture, an embedding-free MLP-Mixer model for on-device NLP that achieves high weight-efficiency thanks to a novel projection layer. We evaluate a pNLP-Mixer model of only one megabyte in size on two multi-lingual semantic parsing datasets, MTOP and multiATIS. Our quantized model achieves 99.4% and 97.8% the performance of mBERT on MTOP and multi-ATIS, while using 170x fewer parameters. Our model consistently beats the state-of-the-art of tiny models (pQRNN), which is twice as large, by a margin up to 7.8% on MTOP., Comment: Accepted at ACL 2023 (industry). 8 pages, 2 figures, 4 tables

The toolbox for material characterization has never been richer than today. Great progress with all kinds of particles and interaction methods provide access to nearly all properties of an object under study. However, a tomographic analysis of the subsurface region remains still a challenge today. In this regard, the Muon-Induced X-ray Emission (MIXE) technique has seen rebirth fueled by the availability of high intensity muon beams. We report here a study conducted at the Paul Scherrer Institute (PSI). It demonstrates that the absence of any beam time-structure leads to low pile-up events and a high signal-to-noise ratio (SNR) with less than one hour acquisition time per sample or data point. This performance creates the perspective to open this technique to a wider audience for the routine investigation of non-destructive and depth-sensitive elemental compositions, for example in rare and precious samples. Using a hetero-structured sample of known elements and thicknesses, we successfully detected the characteristic muonic X-rays, emitted during the capture of a negative muon by an atom, and the gamma-rays resulting from the nuclear capture of the muon, characterizing the capabilities of MIXE at PSI. This sample emphasizes the quality of a continuous beam, and the exceptional SNR at high rates. Such sensitivity will enable totally new statistically intense aspects in the field of MIXE, e.g. elemental 3D-tomography and chemical analysis. Therefore, we are currently advancing our proof-of-concept experiments with the goal of creating a full fledged permanently operated user station to make MIXE available to the wider scientific community as well as industry.

The CREMA collaboration is pursuing a measurement of the ground-state hyperfine splitting (HFS) in muonic hydrogen ($\mu$p) with 1 ppm accuracy by means of pulsed laser spectroscopy to determine the two-photon-exchange contribution with $2\times10^{-4}$ relative accuracy. In the proposed experiment, the $\mu$p atom undergoes a laser excitation from the singlet hyperfine state to the triplet hyperfine state, {then} is quenched back to the singlet state by an inelastic collision with a H$_2$ molecule. The resulting increase of kinetic energy after the collisional deexcitation is used as a signature of a successful laser transition between hyperfine states. In this paper, we calculate the combined probability that a $\mu$p atom initially in the singlet hyperfine state undergoes a laser excitation to the triplet state followed by a collisional-induced deexcitation back to the singlet state. This combined probability has been computed using the optical Bloch equations including the inelastic and elastic collisions. Omitting the decoherence effects caused by {the laser bandwidth and }collisions would overestimate the transition probability by more than a factor of two in the experimental conditions. Moreover, we also account for Doppler effects and provide the matrix element, the saturation fluence, the elastic and inelastic collision rates for the singlet and triplet states, and the resonance linewidth. This calculation thus quantifies one of the key unknowns of the HFS experiment, leading to a precise definition of the requirements for the laser system and to an optimization of the hydrogen gas target where $\mu$p is formed and the laser spectroscopy will occur., Comment: 21 pages, 4 figures

In April 2021, scientists active in muon physics met to discuss and work out the physics case for the new High-Intensity Muon Beams (HIMB) project at PSI that could deliver of order $10^{10}$\,s$^{-1}$ surface muons to experiments. Ideas and concrete proposals were further substantiated over the following months and assembled in the present document. The high intensities will allow for completely new experiments with considerable discovery potential and unique sensitivities. The physics case is outstanding and extremely rich, ranging from fundamental particle physics via chemistry to condensed matter research and applications in energy research and elemental analysis. In all these fields, HIMB will ensure that the facilities S$\mu$S and CHRISP on PSI's High Intensity Proton Accelerator complex HIPA remain world-leading, despite the competition of muon facilities elsewhere., Comment: 116 pages, 42 figures

Recommendations with personalized explanations have been shown to increase user trust and perceived quality and help users make better decisions. Moreover, such explanations allow users to provide feedback by critiquing them. Several algorithms for recommender systems with multi-step critiquing have therefore been developed. However, providing a user-friendly interface based on personalized explanations and critiquing has not been addressed in the last decade. In this paper, we introduce four different web interfaces (available under https://lia.epfl.ch/critiquing/) helping users making decisions and finding their ideal item. We have chosen the hotel recommendation domain as a use case even though our approach is trivially adaptable for other domains. Moreover, our system is model-agnostic (for both recommender systems and critiquing models) allowing a great flexibility and further extensions. Our interfaces are above all a useful tool to help research in recommendation with critiquing. They allow to test such systems on a real use case and also to highlight some limitations of these approaches to find solutions to overcome them., Comment: Accepted at RecSys 2021. 5 pages, 7 figures

Automated predictions require explanations to be interpretable by humans. One type of explanation is a rationale, i.e., a selection of input features such as relevant text snippets from which the model computes the outcome. However, a single overall selection does not provide a complete explanation, e.g., weighing several aspects for decisions. To this end, we present a novel self-interpretable model called ConRAT. Inspired by how human explanations for high-level decisions are often based on key concepts, ConRAT extracts a set of text snippets as concepts and infers which ones are described in the document. Then, it explains the outcome with a linear aggregation of concepts. Two regularizers drive ConRAT to build interpretable concepts. In addition, we propose two techniques to boost the rationale and predictive performance further. Experiments on both single- and multi-aspect sentiment classification tasks show that ConRAT is the first to generate concepts that align with human rationalization while using only the overall label. Further, it outperforms state-of-the-art methods trained on each aspect label independently., Comment: Accepted at ACL 2021 (findings). 15 pages, 10 figures, 7 tables

Recent studies have shown that providing personalized explanations alongside recommendations increases trust and perceived quality. Furthermore, it gives users an opportunity to refine the recommendations by critiquing parts of the explanations. On one hand, current recommender systems model the recommendation, explanation, and critiquing objectives jointly, but this creates an inherent trade-off between their respective performance. On the other hand, although recent latent linear critiquing approaches are built upon an existing recommender system, they suffer from computational inefficiency at inference due to the objective optimized at each conversation's turn. We address these deficiencies with M&Ms-VAE, a novel variational autoencoder for recommendation and explanation that is based on multimodal modeling assumptions. We train the model under a weak supervision scheme to simulate both fully and partially observed variables. Then, we leverage the generalization ability of a trained M&Ms-VAE model to embed the user preference and the critique separately. Our work's most important innovation is our critiquing module, which is built upon and trained in a self-supervised manner with a simple ranking objective. Experiments on four real-world datasets demonstrate that among state-of-the-art models, our system is the first to dominate or match the performance in terms of recommendation, explanation, and multi-step critiquing. Moreover, M&Ms-VAE processes the critiques up to 25.6x faster than the best baselines. Finally, we show that our model infers coherent joint and cross generation, even under weak supervision, thanks to our multimodal-based modeling and training scheme., Comment: Accepted at RecSys 2021. 19 pages, 7 figures, 5 tables