Search

Follow-up conversations with virtual assistants (VAs) enable a user to seamlessly interact with a VA without the need to repeatedly invoke it using a keyword (after the first query). Therefore, accurate Device-directed Speech Detection (DDSD) from the follow-up queries is critical for enabling naturalistic user experience. To this end, we explore the notion of Large Language Models (LLMs) and model the first query when making inference about the follow-ups (based on the ASR-decoded text), via prompting of a pretrained LLM, or by adapting a binary classifier on top of the LLM. In doing so, we also exploit the ASR uncertainty when designing the LLM prompts. We show on the real-world dataset of follow-up conversations that this approach yields large gains (20-40% reduction in false alarms at 10% fixed false rejects) due to the joint modeling of the previous speech context and ASR uncertainty, compared to when follow-ups are modeled alone.

This research explores the modulation of Rayleigh-Plateau Instability (RPI) by adjusting the orientation angle and eccentricity of a wire within a nozzle. We demonstrate that both the angle and eccentricity significantly influence the RPI regimes. They both also influence characteristics, such as velocity, wavelength, and droplet volume. Notably, when wires are both angled and eccentric, the effect of angle prevails. Our approach includes an empirical scaling analysis, comparing gravity, surface tension, and viscosity forces on a single bead, yielding a unified empirical viscous force law, enhancing understanding of RP regime dynamics. This new framework enriches our understanding of the forces at play in RPI and provides practical insights into the manipulation of fluid dynamics in industrial applications.

The $SU(2)_L\times U(1)_Y$ invariance of the Standard Model Effective Field Theory (SMEFT) imposes relations among different low-energy effective field theory Wilson coefficients (WCs), any deviations from which would signal the presence of physics beyond SMEFT. In this work, we investigate two such relations ($C_S = -C_P$, $C_S^\prime = C_P^\prime$) among the scalar and the pseudoscalar new-physics WCs that can contribute to $b\to s \,\tau \tau $ processes. We show that, even when new physics violating these relations would not be measurable in the branching ratios of $B_s \to \tau^+ \tau^-$ and $B\to K^{(*)}\,\tau^+ \tau^-$ at HL-LHC and FCC-ee, it can still manifest itself in the angular distribution of $B\to K^{*0}\,\tau^+ \tau^-$. We identify the combinations of angular observables in this decay channel that are sensitive to scenarios beyond SMEFT. We find that the two observables $S_6^c$ and $A_7$, and their combination, have the potential to identify physics beyond SMEFT., Comment: 10 pages, 9 figures

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

Atmospheric neutrinos can act as a tool to probe the interior of Earth using weak interactions, and can provide information complementary to that obtained from gravitational and seismic measurements. While passing through Earth, multi-GeV neutrinos encounter Earth matter effects due to the coherent forward scattering with the ambient electrons, which alter the neutrino oscillation probabilities. These matter effects depend upon the density distribution of electrons inside Earth, and hence, can be used to determine the internal structure of Earth. In this work, we employ a five-layered model of Earth where the layer densities and radii are modified, keeping the mass and moment of inertia of Earth unchanged and respecting the hydrostatic equilibrium condition. We use the proposed INO-ICAL detector as an example of an atmospheric neutrino experiment that can distinguish between neutrinos and antineutrinos efficiently in the multi-GeV energy range. Our analysis demonstrates the role such an experiment can play in simultaneously constraining the density jumps inside Earth and the location of the core-mantle boundary., Comment: 34 pages, 10 figures, and 3 tables. Comments are welcome

The $SU(2)_L\times U(1)_Y$ invariance of the Standard Model Effective Field Theory (SMEFT) predicts multiple restrictions in the space of Wilson coefficients of $U(1)_{em}$ invariant effective lagrangians such as the Low-energy Effective Field Theory (LEFT), used for low-energy flavor-physics observables, or the Higgs Effective Field Theory (HEFT) in unitary gauge, appropriate for weak-scale observables. In this work, we derive and list all such predictions for semileptonic operators up to dimension 6. We find that these predictions can be expressed as 2223 linear relations among the HEFT/LEFT Wilson coefficients, that are completely independent of any assumptions about the alignment of the mass and flavor bases. These relations connect diverse experimental searches such as rare meson decays, high-$p_T$ dilepton searches, top decays, $Z$-pole observables, charged lepton flavor violating observables and non-standard neutrino interaction searches. We demonstrate how these relations can be used to derive strong indirect constraints on multiple Wilson coefficients that are currently either weakly constrained from direct experiments or have no direct bound at all. These relations also imply, in general, that evidence for new physics in a particular search channel must be accompanied by correlated anomalies in other channels., Comment: 51 pages, 9 figures, 14 tables

Atmospheric neutrinos offer the possibility of exploring the internal structure of Earth. This information is complementary to the traditional probes of seismic and gravitational studies. While propagating through Earth, the multi-GeV neutrinos encounter the Earth's matter effects due to the coherent forward scattering with the ambient electrons, which alters the neutrino oscillation probabilities. We present how well an atmospheric neutrino oscillation experiment like the 50 kt Iron Calorimeter (ICAL) detector at India-based Neutrino Observatory would validate the presence of Earth's core, measure the location of the core-mantle boundary (CMB), and probe the dark matter (DM) inside the Earth in a unique way through Earth matter effects in neutrino oscillations. Owing to good angular resolution, ICAL can observe the core-passing neutrinos efficiently. Due to its magnetized setup, it would be able to observe neutrinos and antineutrinos separately. With 500 kt$\cdot$yr exposure, the presence of Earth's core can be independently confirmed at ICAL with a median $\Delta\chi^2$ of 7.45 (4.83) for normal (inverted) mass ordering. With 1000 kt$\cdot$yr exposure, ICAL would be able to locate the CMB with a precision of about $\pm$ 250 km at $1\sigma$. It would also be sensitive to the possible presence of dark matter with 3.5% of the mass of Earth at $1\sigma$. The charge identification capability of ICAL would play an important role in achieving these precisions., Comment: 4 pages, 2 figures; contribution to proceedings of The European Physical Society Conference on High Energy Physics (EPS-HEP2023)

Coupling of axions or axion-like particles (ALPs) with photons may lead to photons escaping optically opaque regions by oscillating into ALPs. This phenomenon may be viewed as the Light Shining through Wall (LSW) scenario. While this LSW technique has been used previously in controlled laboratory settings to constrain the ALP-photon coupling ($g_{a\gamma}$), we show that this can also be applied in astrophysical environments. We find that obscured magnetars in particular are excellent candidates for this purpose. A fraction of photons emitted by the magnetar may convert to ALPs in the magnetar neighborhood, cross the large absorption column densities, and convert back into photons due to the interstellar magnetic field. Comparing the observed flux with the estimated intrinsic flux from the magnetar, we can constrain the contribution of this process, and hence constrain $g_{a\gamma}$. The effects of resonant conversion near the magnetar as well as ALP-photon oscillations in the interstellar medium are carefully considered. Taking a suitable magnetar candidate PSR J1622-4950, we find that the ALP-photon coupling can be constrained at $g_{a\gamma} \lesssim (10^{-10} - 10^{-11})$ GeV$^{-1}$ for low mass axions ($m_a \lesssim 10^{-12}$ eV). Our study reveals the previously unrealized potential for employing the LSW technique for obscured magnetars for probing and constraining ALP-photon couplings., Comment: 37 pages, 10 figures

Device-directed speech detection (DDSD) is the binary classification task of distinguishing between queries directed at a voice assistant versus side conversation or background speech. State-of-the-art DDSD systems use verbal cues, e.g acoustic, text and/or automatic speech recognition system (ASR) features, to classify speech as device-directed or otherwise, and often have to contend with one or more of these modalities being unavailable when deployed in real-world settings. In this paper, we investigate fusion schemes for DDSD systems that can be made more robust to missing modalities. Concurrently, we study the use of non-verbal cues, specifically prosody features, in addition to verbal cues for DDSD. We present different approaches to combine scores and embeddings from prosody with the corresponding verbal cues, finding that prosody improves DDSD performance by upto 8.5% in terms of false acceptance rate (FA) at a given fixed operating point via non-linear intermediate fusion, while our use of modality dropout techniques improves the performance of these models by 7.4% in terms of FA when evaluated with missing modalities during inference time., Comment: 5 pages

While large language models excel in a variety of natural language processing (NLP) tasks, to perform well on spoken language understanding (SLU) tasks, they must either rely on off-the-shelf automatic speech recognition (ASR) systems for transcription, or be equipped with an in-built speech modality. This work focuses on the former scenario, where LLM's accuracy on SLU tasks is constrained by the accuracy of a fixed ASR system on the spoken input. Specifically, we tackle speech-intent classification task, where a high word-error-rate can limit the LLM's ability to understand the spoken intent. Instead of chasing a high accuracy by designing complex or specialized architectures regardless of deployment costs, we seek to answer how far we can go without substantially changing the underlying ASR and LLM, which can potentially be shared by multiple unrelated tasks. To this end, we propose prompting the LLM with an n-best list of ASR hypotheses instead of only the error-prone 1-best hypothesis. We explore prompt-engineering to explain the concept of n-best lists to the LLM; followed by the finetuning of Low-Rank Adapters on the downstream tasks. Our approach using n-best lists proves to be effective on a device-directed speech detection task as well as on a keyword spotting task, where systems using n-best list prompts outperform those using 1-best ASR hypothesis; thus paving the way for an efficient method to exploit ASR uncertainty via LLMs for speech-based applications., Comment: Added references

For an affine $T$-variety $X$ with the action of a torus $T$, this paper provides a combinatorial description of $X$ with respect to the action of a subtorus $T' \subset T$ in terms of a $T/T'$-invariant pp-divisor. We also describe the corresponding GIT fan.

In the Standard Model Effective Field Theory (SMEFT), the $SU(2)_L\times U(1)_Y$ symmetry of the Standard Model is linearly realized. However, it is possible that more general effective field theories, such as the Higgs Effective Field Theory (HEFT) where this symmetry is realized non-linearly, are needed to describe the data. Identifying physics beyond SMEFT could shed light on the nature of Higgs and the realization of the electroweak symmetry. We explore the possibility of such an identification by studying the effects of scalar and vector new-physics operators on the angular distribution of $\Lambda_b\rightarrow \Lambda_c(\rightarrow\Lambda\pi)\tau\bar\nu_\tau$. This decay is sensitive to the 6-dimensional effective operator $O_V^{LR}\equiv(\bar{\tau}\gamma^\mu P_L\nu_\tau)(\bar{c}\gamma_\mu P_R b)$, which is present in HEFT but suppressed in SMEFT. We identify the angular observables that can have significant contributions from $O_V^{LR}$, and hence would be useful for probing not only BSM physics but also physics beyond SMEFT. We further find that constraining the branching ratio of $B_c\rightarrow \tau \bar \nu_\tau$ would be crucial for performing this task., Comment: 17 pages, 9 figures, 5 tables. Major changes from version 1: updated NP parameter fits, effects of a stronger bound on semileptonic Bc decay included, sections rearranged, references updated. Version to be published in PRD

The paper provides a computation of the equivariant Chow group of a rational, complete, complexity one $T$-variety

The quantum nature of neutrino oscillations would be reflected in the mismatch between the neutrino survival probabilities with and without an intermediate observation. We propose this ``quantum mismatch'' as a measure of quantumness in neutrino oscillations, which precisely extracts the interference term in the two-flavor limit. In the full three-flavor scenario, we provide modified definitions of the Leggett-Garg and quantum mismatch measures. These are applicable for long-baseline and reactor neutrino experiments that measure neutrino survival probabilities with negligible matter effects., Comment: 6 pages, 4 figures

Objective: Reflex testing protocols allow clinical laboratories to perform second line diagnostic tests on existing specimens based on the results of initially ordered tests. Reflex testing can support optimal clinical laboratory test ordering and diagnosis. In current clinical practice, reflex testing typically relies on simple "if-then" rules; however, this limits their scope since most test ordering decisions involve more complexity than a simple rule will allow. Here, using the analyte ferritin as an example, we propose an alternative machine learning-based approach to "smart" reflex testing with a wider scope and greater impact than traditional rule-based approaches. Methods: Using patient data, we developed a machine learning model to predict whether a patient getting CBC testing will also have ferritin testing ordered, consider applications of this model to "smart" reflex testing, and evaluate the model by comparing its performance to possible rule-based approaches. Results: Our underlying machine learning models performed moderately well in predicting ferritin test ordering and demonstrated greater suitability to reflex testing than rule-based approaches. Using chart review, we demonstrate that our model may improve ferritin test ordering. Finally, as a secondary goal, we demonstrate that ferritin test results are missing not at random (MNAR), a finding with implications for unbiased imputation of missing test results. Conclusions: Machine learning may provide a foundation for new types of reflex testing with enhanced benefits for clinical diagnosis and laboratory utilization management.

Atmospheric neutrinos provide a unique avenue to explore the internal structure of Earth based on weak interactions, which is complementary to seismic studies and gravitational measurements. In this work, we demonstrate that the atmospheric neutrino oscillations in the presence of Earth matter can serve as an important tool to locate the core-mantle boundary (CMB). An atmospheric neutrino detector like the proposed 50 kt magnetized ICAL at INO can observe the core-passing neutrinos efficiently. These neutrinos would have experienced the MSW resonance and the parametric or neutrino oscillation length resonance. The net effect of these resonances on neutrino flavor conversions depends upon the location of CMB and the density jump at that radius. We quantify the capability of ICAL to measure the location of CMB in the context of multiple three-layered models of Earth. For the model where the density and the radius of core are kept flexible while the mass and radius of Earth as well as the densities of outer and inner mantle are fixed, ICAL can determine the location of CMB with a 1$\sigma$ precision of about 250 km with an exposure of 1000 kt$\cdot$yr. With the 81-layered PREM profile, this $1\sigma$ precision would be about 350 km. The charge identification capability of ICAL plays an important role in achieving this precision., Comment: 36 pages, 11 figures, 5 tables, and two appendices. A new appendix B with a new figure 11 added to show the sensitivities using the 81-layered PREM profile and some sentences added in the abstract, main text, and summary to discuss these new results. The results obtained without marginalizing over the oscillation parameters in the fit added in table 5. Matches with the published version in JHEP

We analytically calculate the neutrino conversion probability $P_{\mu e}$ in the presence of sterile neutrinos, with exact dependence on $\Delta m^2_{41}$ and with matter effects explicitly included. Using perturbative expansion in small parameters, the terms involving the small mixing angles $\theta_{24}$ and $\theta_{34}$ can be separated out, with $\theta_{34}$ dependence only arising due to matter effects. We express $P_{\mu e}$ in terms of the quantities of the form $\sin(x)/x$, which helps in elucidating its dependence on matter effects and a wide range of $\Delta m^2_{41}$ values. Our analytic expressions allow us to predict the effects of the sign of $\Delta m^2_{41}$ at a long baseline experiment like DUNE. We numerically calculate the sensitivity of DUNE to the sterile mass ordering and find that this sensitivity can be significant in the range $|\Delta m^2_{41}| \sim (10^{-4} - 10^{-2})$ eV$^2$, for either mass ordering of active neutrinos. The dependence of this sensitivity on the value of $\Delta m^2_{41}$ for all mass ordering combinations can be explained by investigating the resonance-like terms appearing due to the interplay between the sterile sector and matter effects., Comment: 25 pages, 5 figures

Accurate prediction of the user intent to interact with a voice assistant (VA) on a device (e.g. on the phone) is critical for achieving naturalistic, engaging, and privacy-centric interactions with the VA. To this end, we present a novel approach to predict the user's intent (the user speaking to the device or not) directly from acoustic and textual information encoded at subword tokens which are obtained via an end-to-end ASR model. Modeling directly the subword tokens, compared to modeling of the phonemes and/or full words, has at least two advantages: (i) it provides a unique vocabulary representation, where each token has a semantic meaning, in contrast to the phoneme-level representations, (ii) each subword token has a reusable "sub"-word acoustic pattern (that can be used to construct multiple full words), resulting in a largely reduced vocabulary space than of the full words. To learn the subword representations for the audio-to-intent classification, we extract: (i) acoustic information from an E2E-ASR model, which provides frame-level CTC posterior probabilities for the subword tokens, and (ii) textual information from a pre-trained continuous bag-of-words model capturing the semantic meaning of the subword tokens. The key to our approach is the way it combines acoustic subword-level posteriors with text information using the notion of positional-encoding in order to account for multiple ASR hypotheses simultaneously. We show that our approach provides more robust and richer representations for audio-to-intent classification, and is highly accurate with correctly mitigating 93.3% of unintended user audio from invoking the smart assistant at 99% true positive rate.

Precision measurements of neutrino oscillation parameters have provided a tremendous boost to the search for sub-leading effects due to several beyond the Standard Model scenarios in neutrino oscillation experiments. Among these, two of the well-studied scenarios are Lorentz violation (LV) and non-standard interactions (NSI), both of which can affect neutrino oscillations significantly. We point out that, at a long-baseline experiment where the neutrino oscillation probabilities can be well-approximated by using the line-averaged constant matter density, the effects of these two scenarios can mimic each other. This would allow the limits obtained at such an experiment on one of the above scenarios to be directly translated to the limits on the other scenario. However, for the same reason, it would be difficult to distinguish between LV and NSI at a long-baseline experiment. We show that the observations of atmospheric neutrinos, which travel a wide range of baselines and may encounter sharp density changes at the core-mantle boundary, can break this degeneracy. We observe that identifying neutrinos and antineutrinos separately, as can be done at INO-ICAL, can enhance the capability of atmospheric neutrino experiments to discriminate between these two new-physics scenarios., Comment: 13 pages, 6 figures, 1 appendix. The title changed, an explanatory appendix added, the presentation of figure 6 changed (results unchanged). This version has the same contents as the published article in Physics Letters B

We present compact analytic expressions for 3-flavor neutrino oscillation probabilities with invisible neutrino decay, where matter effects have been explicitly included. We take into account the possibility that the oscillation and decay components of the effective Hamiltonian do not commute. This is achieved by employing the techniques of inverse Baker-Campbell-Hausdorff (BCH) expansion and the Cayley-Hamilton theorem applied in the 3-flavor framework. If only the vacuum mass eigenstate $\nu_3$ decays, we show that the treatment of neutrino propagation may be reduced to an effective 2-flavor analysis in the One Mass Scale Dominance (OMSD) approximation. The oscillation probabilities for $P_{\mu\mu}$, $P_{ee}$, $P_{e\mu}$ and $P_{\mu e}$ -- relevant for reactor, long baseline and atmospheric neutrino experiments -- are obtained as perturbative expansions for the case of only $\nu_3$ decay, as well as for the more general scenario where all components of the decay matrix are non-zero. The analytic results thus obtained match the exact numerical results for constant density matter to a high precision and provide physical insights into possible effects of the decay of neutrinos as they propagate through Earth matter. We find that the effects of neutrino decay are most likely to be observable in $P_{\mu\mu}$. We also point out that at any long baseline, the oscillation dips in $P_{\mu\mu}$ can show higher survival probabilities in the case with decay than without decay, and explain this feature using our analytic approximations., Comment: 35 pages, 6 figures, references updated in v2 (no other changes)