Search for: All records

Current approaches to novelty or anomaly detection are based on deep neural networks. Despite their effectiveness, neural networks are also vulnerable to imperceptible deformations of the input data. This is a serious issue in critical applications, or when data alterations are generated by an adversarial attack. While this is a known problem that has been studied in recent years for the case of supervised learn- ing, the case of novelty detection has received very limited attention. Indeed, in this latter setting the learning is typically unsupervised because outlier data is not available during training, and new approaches for this case need to be investigated. We propose a new prior that aims at learning a robust likelihood for the novelty test, as a defense against attacks. We also integrate the same prior with a state-of-the- art novelty detection approach. Because of the geometric properties of that approach, the resulting robust training is computationally very efficient. An initial evaluation of the method indicates that it is effective at improving performance with respect to the standard models in the absence and presence of attacks. 

Quantum error correction can preserve quantum information in the presence of local errors, but correlated errors are fatal. For superconducting qubits, high-energy particle impacts from background radioactivity produce energetic phonons that travel throughout the substrate and create excitations above the superconducting ground state, known as quasiparticles, which can poison all qubits on the chip. We use normal metal reservoirs on the chip back side to downconvert phonons to low energies where they can no longer poison qubits. We introduce a pump-probe scheme involving controlled injection of pair-breaking phonons into the qubit chips. We examine quasiparticle poisoning on chips with and without back-side metallization and demonstrate a reduction in the flux of pair-breaking phonons by over a factor of 20. We use a Ramsey interferometer scheme to simultaneously monitor quasiparticle parity on three qubits for each chip and observe a two-order of magnitude reduction in correlated poisoning due to background radiation.

 

The BigCode community, an open-scientific collaboration working on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder and StarCoderBase: 15.5B parameter models with 8K context length, infilling capabilities and fast large-batch inference enabled by multi-query attention. StarCoderBase is trained on 1 trillion tokens sourced from The Stack, a large collection of permissively licensed GitHub repositories with inspection tools and an opt-out process. We fine-tuned StarCoderBase on 35B Python tokens, resulting in the creation of StarCoder. We perform the most comprehensive evaluation of Code LLMs to date and show that StarCoderBase outperforms every open Code LLM that supports multiple programming languages and matches or outperforms the OpenAI code-cushman-001 model. Furthermore, StarCoder outperforms every model that is fine-tuned on Python, can be prompted to achieve 40% pass@1 on HumanEval, and still retains its performance on other programming languages. We take several important steps towards a safe open-access model release, including an improved PII redaction pipeline and a novel attribution tracing tool, and make the StarCoder models publicly available under a more commercially viable version of the Open Responsible AI Model license. 

Optically active defects in 2D materials, such as hexagonal boron nitride (hBN) and transition metal dichalcogenides (TMDs), are an attractive class of single-photon emitters with high brightness, room-temperature operation, site-specific engineering of emitter arrays, and tunability with external strain and electric fields. In this work, we demonstrate a novel approach to precisely align and embed hBN and TMDs within background-free silicon nitride microring resonators. Through the Purcell effect, high-purity hBN emitters exhibit a cavity-enhanced spectral coupling efficiency up to 46% at room temperature, which exceeds the theoretical limit for cavity-free waveguide-emitter coupling and previous demonstrations by nearly an order-of-magnitude. The devices are fabricated with a CMOS-compatible process and exhibit no degradation of the 2D material optical properties, robustness to thermal annealing, and 100 nm positioning accuracy of quantum emitters within single-mode waveguides, opening a path for scalable quantum photonic chips with on-demand single-photon sources. 

Background Cerebellar electrical stimulation has shown promise in improving motor recovery post-stroke in both rodent and human studies. Past studies have used motor evoked potentials (MEPs) to evaluate how cerebellar stimulation modulates ongoing activity in the cortex, but the underlying mechanisms are incompletely understood. Here we used invasive electrophysiological recordings from the intact and stroke-injured rodent primary motor cortex (M1) to assess how epidural cerebellar stimulation modulates neural dynamics at the level of single neurons as well as at the level of mesoscale dynamics. Methods We recorded single unit spiking and local field potentials (LFPs) in both the intact and acutely stroke-injured M1 contralateral to the stimulated cerebellum in adult Long-Evans rats under anesthesia. We analyzed changes in the firing rates of single units, the extent of synchronous spiking and power spectral density (PSD) changes in LFPs during and post-stimulation. Results Our results show that post-stimulation, the firing rates of a majority of M1 neurons changed significantly with respect to their baseline rates. These firing rate changes were diverse in character, as the firing rate of some neurons increased while others decreased. Additionally, these changes started to set in during stimulation. Furthermore, cross-correlation analysis showed a significant increase in coincident firing amongst neuronal pairs. Interestingly, this increase in synchrony was unrelated to the direction of firing rate change. We also found that neuronal ensembles derived through principal component analysis were more active post-stimulation. Lastly, these changes occurred without a significant change in the overall spectral power of LFPs post-stimulation. Conclusions Our results show that cerebellar stimulation caused significant, long-lasting changes in the activity patterns of M1 neurons by altering firing rates, boosting neural synchrony and increasing neuronal assemblies’ activation strength. Our study provides evidence that cerebellar stimulation can directly modulate cortical dynamics. Since these results are present in the perilesional cortex, our data might also help explain the facilitatory effects of cerebellar stimulation post-stroke. 

Abstract The ground-based gamma-ray observatory Very Energetic Radiation Imaging Telescope Array System (VERITAS, https://veritas.sao.arizona.edu/ ) is sensitive to photons of astrophysical origin with energies in the range between ≈85 GeV and ≈30 TeV. The instrument consists of four 12 m diameter imaging Cherenkov telescopes operating at the Fred Lawrence Whipple Observatory in southern Arizona. VERITAS started four-telescope operations in 2007 and collects about 1100 hr of good-weather data per year. The VERITAS collaboration has published over 100 journal articles since 2008 reporting on gamma-ray observations of a large variety of objects: Galactic sources like supernova remnants, pulsar wind nebulae, and binary systems; extragalactic sources like star-forming galaxies, dwarf-spheroidal galaxies, and highly variable active galactic nuclei. This note presents VTSCat: the catalog of high-level data products from all VERITAS publications.