Search

Precision optical measurements of the electron-spin precession of nitrogen-vacancy (NV) centers in diamond form the basis of numerous applications. The most sensitivity-demanding applications, such as femtotesla magnetometry, require the ability to measure changes in GHz spin transition frequencies at the sub-millihertz level, corresponding to a fractional resolution of better than 10^{-12}. Here we study the impact of microwave (MW) phase noise on the response of an NV sensor. Fluctuations of the phase of the MW waveform cause undesired rotations of the NV spin state. These fluctuations are imprinted in the optical readout signal and, left unmitigated, are indistinguishable from magnetic field noise. We show that the phase noise of several common commercial MW generators results in an effective pT s^{1/2}-range noise floor that varies with the MW carrier frequency and the detection frequency of the pulse sequence. The data are described by a frequency domain model incorporating the MW phase noise spectrum and the filter-function response of the sensing protocol. For controlled injection of white and random-walk phase noise, the observed NV magnetic noise floor is described by simple analytic expressions that accurately capture the scaling with pulse sequence length and the number of pi pulses. We outline several strategies to suppress the impact of MW phase noise and implement a version, based on gradiometry, that realizes a >10-fold suppression. Our study highlights an important challenge in the pursuit of sensitive diamond quantum sensors and is applicable to other qubit systems with a large transition frequency., Comment: Main text: 8 pages, 5 figures. Entire manuscript including Appendices: 22 pages, 14 figures, 6 tables

Preference-based reinforcement learning (PbRL) aligns a robot behavior with human preferences via a reward function learned from binary feedback over agent behaviors. We show that dynamics-aware reward functions improve the sample efficiency of PbRL by an order of magnitude. In our experiments we iterate between: (1) learning a dynamics-aware state-action representation (z^{sa}) via a self-supervised temporal consistency task, and (2) bootstrapping the preference-based reward function from (z^{sa}), which results in faster policy learning and better final policy performance. For example, on quadruped-walk, walker-walk, and cheetah-run, with 50 preference labels we achieve the same performance as existing approaches with 500 preference labels, and we recover 83\% and 66\% of ground truth reward policy performance versus only 38\% and 21\%. The performance gains demonstrate the benefits of explicitly learning a dynamics-aware reward model. Repo: \texttt{https://github.com/apple/ml-reed}., Comment: CoRL 2023. arXiv admin note: substantial text overlap with arXiv:2211.06527

Self-supervised features are typically used in place of filter-bank features in speaker verification models. However, these models were originally designed to ingest filter-bank features as inputs, and thus, training them on top of self-supervised features assumes that both feature types require the same amount of learning for the task. In this work, we observe that pre-trained self-supervised speech features inherently include information required for downstream speaker verification task, and therefore, we can simplify the downstream model without sacrificing performance. To this end, we revisit the design of the downstream model for speaker verification using self-supervised features. We show that we can simplify the model to use 97.51% fewer parameters while achieving a 29.93% average improvement in performance on SUPERB. Consequently, we show that the simplified downstream model is more data efficient compared to baseline--it achieves better performance with only 60% of the training data.

This paper introduces ESPnet-SPK, a toolkit designed with several objectives for training speaker embedding extractors. First, we provide an open-source platform for researchers in the speaker recognition community to effortlessly build models. We provide several models, ranging from x-vector to recent SKA-TDNN. Through the modularized architecture design, variants can be developed easily. We also aspire to bridge developed models with other domains, facilitating the broad research community to effortlessly incorporate state-of-the-art embedding extractors. Pre-trained embedding extractors can be accessed in an off-the-shelf manner and we demonstrate the toolkit's versatility by showcasing its integration with two tasks. Another goal is to integrate with diverse self-supervised learning features. We release a reproducible recipe that achieves an equal error rate of 0.39% on the Vox1-O evaluation protocol using WavLM-Large with ECAPA-TDNN., Comment: 5 pages, 3 figures, 7 tables, Interspeech 2024

Fully-test-time adaptation (F-TTA) can mitigate performance loss due to distribution shifts between train and test data (1) without access to the training data, and (2) without knowledge of the model training procedure. In online F-TTA, a pre-trained model is adapted using a stream of test samples by minimizing a self-supervised objective, such as entropy minimization. However, models adapted with online using entropy minimization, are unstable especially in single sample settings, leading to degenerate solutions, and limiting the adoption of TTA inference strategies. Prior works identify noisy, or unreliable, samples as a cause of failure in online F-TTA. One solution is to ignore these samples, which can lead to bias in the update procedure, slow adaptation, and poor generalization. In this work, we present a general framework for improving robustness of F-TTA to these noisy samples, inspired by self-paced learning and robust loss functions. Our proposed approach, Robust Entropy Adaptive Loss Minimization (REALM), achieves better adaptation accuracy than previous approaches throughout the adaptation process on corruptions of CIFAR-10 and ImageNet-1K, demonstrating its effectiveness., Comment: Accepted at WACV 2024, 17 pages, 7 figures, 11 tables

We present Spatial LibriSpeech, a spatial audio dataset with over 650 hours of 19-channel audio, first-order ambisonics, and optional distractor noise. Spatial LibriSpeech is designed for machine learning model training, and it includes labels for source position, speaking direction, room acoustics and geometry. Spatial LibriSpeech is generated by augmenting LibriSpeech samples with 200k+ simulated acoustic conditions across 8k+ synthetic rooms. To demonstrate the utility of our dataset, we train models on four spatial audio tasks, resulting in a median absolute error of 6.60{\deg} on 3D source localization, 0.43m on distance, 90.66ms on T30, and 2.74dB on DRR estimation. We show that the same models generalize well to widely-used evaluation datasets, e.g., obtaining a median absolute error of 12.43{\deg} on 3D source localization on TUT Sound Events 2018, and 157.32ms on T30 estimation on ACE Challenge.

Quantum sensors offer unparalleled precision, accuracy, and sensitivity for a variety of measurement applications. We report a compact magnetometer based on a ferrimagnetic sensing element in an oscillator architecture that circumvents challenges common to other quantum sensing approaches such as limited dynamic range, limited bandwidth, and dependence on vacuum, cryogenic, or laser components. The device exhibits a fixed, calibration-free response governed by the electron gyromagnetic ratio. Exchange narrowing in the ferrimagnetic material produces sub-MHz transition linewidths despite the high unpaired spin density ($\sim 10^{22}$ cm$^{-3}$). The magnetometer achieves a minimum sensitivity of 100 fT/$\sqrt{\text{Hz}}$ to AC magnetic fields of unknown phase and a sensitivity below 200 fT/$\sqrt{\text{Hz}}$ over a bandwidth $\gtrsim \! 1$ MHz. By encoding magnetic field in frequency rather than amplitude, the device provides a dynamic range in excess of 1 mT. The passive, thermal initialization of the sensor's quantum state requires only a magnetic bias field, greatly reducing power requirements compared to laser-initialized quantum sensors. With additional development, this device promises to be a leading candidate for high-performance magnetometry outside the laboratory, and the oscillator architecture is expected to provide advantages across a wide range of sensing platforms., Comment: 18 pages, 4 figures

Quantum sensing with solid-state spins offers the promise of high spatial resolution, bandwidth, and dynamic range at sensitivities comparable to more mature quantum sensing technologies, such as atomic vapor cells and superconducting devices. However, despite comparable theoretical sensitivity limits, the performance of bulk solid-state quantum sensors has so far lagged behind these more mature alternatives. A recent review~\cite{barry2020sensitivity} suggests several paths to improve performance of magnetometers employing nitrogen-vacancy defects in diamond, the most-studied solid-state quantum sensing platform. Implementing several suggested techniques, we demonstrate the most sensitive nitrogen-vacancy-based bulk magnetometer reported to date. Our approach combines tailored diamond growth to achieve low strain and long intrinsic dephasing times, the use of double-quantum Ramsey and Hahn echo magnetometry sequences for broadband and narrowband magnetometry respectively, and P1 driving to further extend dephasing time. Notably, the device does not include a flux concentrator, preserving the fixed response of the NVs to magnetic field. The magnetometer realizes a broadband \textcolor{mhsnew}{near-}DC sensitivity $\sim 460$~fT$\cdot$s$^{1/2}$ and a narrowband AC sensitivity $\sim 210$~fT$\cdot$s$^{1/2}$. We describe the experimental setup in detail and highlight potential paths for future improvement.

Multi-Exit models (MEMs) use an early-exit strategy to improve the accuracy and efficiency of deep neural networks (DNNs) by allowing samples to exit the network before the last layer. However, the effectiveness of MEMs in the presence of distribution shifts remains largely unexplored. Our work examines how distribution shifts generated by common image corruptions affect the accuracy/efficiency of MEMs. We find that under common corruptions, early-exiting at the first correct exit reduces the inference cost and provides a significant boost in accuracy ( 10%) over exiting at the last layer. However, with realistic early-exit strategies, which do not assume knowledge about the correct exits, MEMs still reduce inference cost but provide a marginal improvement in accuracy (1%) compared to exiting at the last layer. Moreover, the presence of distribution shift widens the gap between an MEM's maximum classification accuracy and realistic early-exit strategies by 5% on average compared with the gap on in-distribution data. Our empirical analysis shows that the lack of calibration due to a distribution shift increases the susceptibility of such early-exit strategies to exit early and increases misclassification rates. Furthermore, the lack of calibration increases the inconsistency in the predictions of the model across exits, leading to both inefficient inference and more misclassifications compared with evaluation on in-distribution data. Finally, we propose two metrics, underthinking and overthinking, that quantify the different behavior of practical early-exit strategy under distribution shifts, and provide insights into improving the practical utility of MEMs., Comment: 16 pages, 22 figures

Preference-based reinforcement learning (RL) algorithms help avoid the pitfalls of hand-crafted reward functions by distilling them from human preference feedback, but they remain impractical due to the burdensome number of labels required from the human, even for relatively simple tasks. In this work, we demonstrate that encoding environment dynamics in the reward function (REED) dramatically reduces the number of preference labels required in state-of-the-art preference-based RL frameworks. We hypothesize that REED-based methods better partition the state-action space and facilitate generalization to state-action pairs not included in the preference dataset. REED iterates between encoding environment dynamics in a state-action representation via a self-supervised temporal consistency task, and bootstrapping the preference-based reward function from the state-action representation. Whereas prior approaches train only on the preference-labelled trajectory pairs, REED exposes the state-action representation to all transitions experienced during policy training. We explore the benefits of REED within the PrefPPO [1] and PEBBLE [2] preference learning frameworks and demonstrate improvements across experimental conditions to both the speed of policy learning and the final policy performance. For example, on quadruped-walk and walker-walk with 50 preference labels, REED-based reward functions recover 83% and 66% of ground truth reward policy performance and without REED only 38\% and 21\% are recovered. For some domains, REED-based reward functions result in policies that outperform policies trained on the ground truth reward.

Human skeleton point clouds are commonly used to automatically classify and predict the behaviour of others. In this paper, we use a contrastive self-supervised learning method, SimCLR, to learn representations that capture the semantics of skeleton point clouds. This work focuses on systematically evaluating the effects that different algorithmic decisions (including augmentations, dataset partitioning and backbone architecture) have on the learned skeleton representations. To pre-train the representations, we normalise six existing datasets to obtain more than 40 million skeleton frames. We evaluate the quality of the learned representations with three downstream tasks: skeleton reconstruction, motion prediction, and activity classification. Our results demonstrate the importance of 1) combining spatial and temporal augmentations, 2) including additional datasets for encoder training, and 3) and using a graph neural network as an encoder., Comment: 8 pages, 2 figures, 4 tables. Accepted at the NeurIPS 2022 Workshop: Self-Supervised Learning - Theory and Practice

Synthesizing natural head motion to accompany speech for an embodied conversational agent is necessary for providing a rich interactive experience. Most prior works assess the quality of generated head motion by comparing them against a single ground-truth using an objective metric. Yet there are many plausible head motion sequences to accompany a speech utterance. In this work, we study the variation in the perceptual quality of head motions sampled from a generative model. We show that, despite providing more diverse head motions, the generative model produces motions with varying degrees of perceptual quality. We finally show that objective metrics commonly used in previous research do not accurately reflect the perceptual quality of generated head motions. These results open an interesting avenue for future work to investigate better objective metrics that correlate with human perception of quality., Comment: Submitted to ICASSP 2023

Quantum sensing of low-frequency magnetic fields using nitrogen-vacancy (NV) center ensembles has been demonstrated in multiple experiments with sensitivities as low as $\sim$1 pT/$\sqrt{\text{Hz}}$. To date, however, demonstrations of high-frequency magnetometry in the GHz regime with NV diamond are orders of magnitude less sensitive, above the nT/$\sqrt{\text{Hz}}$ level. Here we adapt for microwave frequencies techniques that have enabled high-performance, low-frequency quantum sensors. Using a custom-grown NV-enriched diamond combined with a noise cancellation scheme designed for high-frequency sensing, we demonstrate a Rabi-sequence-based magnetometer able to detect microwave fields near 2.87 GHz with a record sensitivity of 3.4 pT/$\sqrt{\textrm{Hz}}$. We demonstrate both amplitude and phase sensing and project tunability over a 300 MHz frequency range. This result increases the viability of NV ensembles to serve as microwave circuitry imagers and near-field probes of antennas., Comment: 7 pages, 4 figures

Generating realistic lip motion from audio to simulate speech production is critical for driving natural character animation. Previous research has shown that traditional metrics used to optimize and assess models for generating lip motion from speech are not a good indicator of subjective opinion of animation quality. Devising metrics that align with subjective opinion first requires understanding what impacts human perception of quality. In this work, we focus on the degree of articulation and run a series of experiments to study how articulation strength impacts human perception of lip motion accompanying speech. Specifically, we study how increasing under-articulated (dampened) and over-articulated (exaggerated) lip motion affects human perception of quality. We examine the impact of articulation strength on human perception when considering only lip motion, where viewers are presented with talking faces represented by landmarks, and in the context of embodied characters, where viewers are presented with photo-realistic videos. Our results show that viewers prefer over-articulated lip motion consistently more than under-articulated lip motion and that this preference generalizes across different speakers and embodiments., Comment: Submitted to ICASSP 2023

Federated learning enables the deployment of machine learning to problems for which centralized data collection is impractical. Adding differential privacy guarantees bounds on privacy while data are contributed to a global model. Adding personalization to federated learning introduces new challenges as we must account for preferences of individual users, where a data sample could have conflicting labels because one sub-population of users might view an input positively, but other sub-populations view the same input negatively. We present FedEmbed, a new approach to private federated learning for personalizing a global model that uses (1) sub-populations of similar users, and (2) personal embeddings. We demonstrate that current approaches to federated learning are inadequate for handling data with conflicting labels, and we show that FedEmbed achieves up to 45% improvement over baseline approaches to personalized private federated learning., Comment: 15 pages

Quantum sensors based on spin defect ensembles have seen rapid development in recent years, with a wide array of target applications. Historically, these sensors have used optical methods to prepare or read out quantum states. However, these methods are limited to optically-polarizable spin defects, and the spin ensemble size is typically limited by the available optical power or acceptable optical heat load. We demonstrate a solid-state sensor employing a non-optical state preparation technique, which harnesses thermal population imbalances induced by the defect's zero-field splitting. Readout is performed using the recently-demonstrated microwave cavity readout technique, resulting in a sensor architecture that is entirely non-optical and broadly applicable to all solid-state paramagnetic defects with a zero-field splitting. The implementation in this work uses Cr$^{3+}$ defects in a sapphire (Al$_2$O$_3$) crystal and a simple microwave architecture where the host crystal also serves as the high quality-factor resonator. This approach yields a near-unity filling factor and high single-spin-photon coupling, producing a magnetometer with a broadband sensitivity of 9.7 pT/$\sqrt{\text{Hz}}$., Comment: 18 pages, 10 figures