Dynamic states offer extended possibilities to control the properties of quantum matter. Recent efforts are focused on studying the ordered states which appear exclusively under the time-dependent drives. Here, we demonstrate a class of systems with dynamic spin-triplet superconducting order stimulated by the alternating electric field. The effect is based on the interplay of ferromagnetism, interfacial spin-orbital coupling, and the condensate motion driven by the field, which converts hidden static p -wave order produced by the joint action of the ferromagnetism and the spin-orbital coupling into dynamic s-wave equal-spin-triplet correlations. We demonstrate that the critical current of Josephson junctions hosting these states is proportional to the electromagnetic power supplied by external irradiation or the ac current source. Based on these unusual properties, we propose the scheme of a Josephson transistor which can be switched by the ac voltage and demonstrates an even-numbered sequence of Shapiro steps. Combining the photoactive Josephson junctions with recently discovered Josephson phase batteries, we find photomagnetic SQUID devices that can generate spontaneous magnetic fields while exposed to irradiation. less

Representation learning algorithms offer the opportunity to learn invariant representations of the input data with regard to nuisance factors. Many authors have leveraged such strategies to learn fair representations, i.e., vectors where information about sensitive attributes is removed. These methods are attractive as they may be interpreted as minimizing the mutual information between a neural layer's activations and a sensitive attribute. However, the theoretical grounding of such methods relies either on the computation of infinitely accurate adversaries or on minimizing a variational upper bound of a mutual information estimate. In this paper, we propose a methodology for direct computation of the mutual information between a neural layer and a sensitive attribute. We employ stochastically-activated binary neural networks, which lets us treat neurons as random variables. We are then able to compute (not bound) the mutual information between a layer and a sensitive attribute and use this information as a regularization factor during gradient descent. We show that this method compares favorably with the state of the art in fair representation learning and that the learned representations display a higher level of invariance compared to full-precision neural networks. less

ScienceCast is a free, open-access platform that links papers and video-abstracts. It also provides a feed of research videos tailored to a reader's interests, and a discussion forum to interact with peers and authors. less

We show how strange metallic behavior emerges from a simple effective semi-holographic theory with only two couplings. This model is justified by the Wilsonian RG approach but ignores order parameters, and describes carrier electrons interacting with a critical sector with only two effective dimensionless couplings. We establish that this model has an emergent universal spectral function near the Fermi surface at an optimal ratio of the two effective couplings when the critical exponent ν lies between 0.66 and 0.95 for a wide range of temperatures which could be as low as 1 percent of the Fermi energy when ν is at the higher end. The linear in T resistivity and quadratic in T Hall resistivity (the latter for a spherical Fermi surface) follow from the universality of the spectral function for a wide range of temperatures in which the relevant loop integrals get contributions mainly from momenta near the Fermi surface. Our spectral functions fit very well with data from ARPES experiments for optimal, over and under-doping with a fixed exponent over a range of temperatures, validating universality at optimal doping. We obtain a refined Planckian dissipation picture where the scattering time τ≈f⋅ℏ/(kBT) with f almost independent of all model parameters except for the overall strength of the two couplings when their ratio is optimal. We find f≈1 when the couplings take maximal value for which strange metallic behavior is exhibited. However, f≈10 when the couplings are smaller. Although not derivable from quasi-particles, our results at optimal doping fit a Drude-type phenomenology with Planckian scattering time, and an almost model-independent n/m, which is about 2π times the Fermi liquid value at same EF. less

From natural language processing to genome sequencing, large-scale machine learning models are bringing advances to a broad range of fields. Many of these models are too large to be trained on a single machine, and instead must be distributed across multiple devices. This has motivated the research of new compute and network systems capable of handling such tasks. In particular, recent work has focused on developing management schemes which decide how to allocate distributed resources such that some overall objective, such as minimising the job completion time (JCT), is optimised. However, such studies omit explicit consideration of how much a job should be distributed, usually assuming that maximum distribution is desirable. In this work, we show that maximum parallelisation is sub-optimal in relation to user-critical metrics such as throughput and blocking rate. To address this, we propose PAC-ML (partitioning for asynchronous computing with machine learning). PAC-ML leverages a graph neural network and reinforcement learning to learn how much to partition computation graphs such that the number of jobs which meet arbitrary user-defined JCT requirements is maximised. In experiments with five real deep learning computation graphs on a recently proposed optical architecture across four user-defined JCT requirement distributions, we demonstrate PAC-ML achieving up to 56.2% lower blocking rates in dynamic job arrival settings than the canonical maximum parallelisation strategy used by most prior works. less

Non-negative tensor factorization models enable predictive analysis on count data. Among them, Bayesian Poisson-Gamma models can derive full posterior distributions of latent factors and are less sensitive to sparse count data. However, current inference methods for these Bayesian models adopt restricted update rules for the posterior parameters. They also fail to share the update information to better cope with the data sparsity. Moreover, these models are not endowed with a component that handles the imbalance in count data values. In this paper, we propose a novel variational auto-encoder framework called VAE-BPTF which addresses the above issues. It uses multi-layer perceptron networks to encode and share complex update information. The encoded information is then reweighted per data instance to penalize common data values before aggregated to compute the posterior parameters for the latent factors. Under synthetic data evaluation, VAE-BPTF tended to recover the right number of latent factors and posterior parameter values. It also outperformed current models in both reconstruction errors and latent factor (semantic) coherence across five real-world datasets. Furthermore, the latent factors inferred by VAE-BPTF are perceived to be meaningful and coherent under a qualitative analysis. less

Welcome to ScienceCast - a universal ecosystem for research, publishing, and exchange of scientific ideas using 21st century tools. ScienceCast was founded by a group of scientists and experts in academic publishing in response to a major problem facing modern day science - exponential proliferation of scientific data. The problem is not handling the big data, but extracting gems and useful insights from the avalanche of scientific papers. Just 20 years ago, it was possible for researchers to follow all developments in their discipline, but information overload makes the traditional modes of research dissemination no longer sustainable. To partially overcome this challenge, ScienceCast introduces a new model, whose key component is a stream of short video pitches summarizing a scientific discovery with links to the full-length articles & data. It also provides space for professional research discussion and continuous peer feedback. The scientific content is reviewed by a professional scientific board and select contributions will be disseminated broadly to the relevant communities, journal editors, funding agencies, and the public. To submit a contribution simply record a short 3-5 minute video on your research (using zoom or a built-in screen recording functionality, e.g. Shift-Command-5 on Mac), go over your manuscript or 2-3 slides highlighting the main results of your work, and provide a link to the full-text of your paper for more details. less

Trust in science increases when scientists and the outlets certifying their work honor science’s norms. Scientists often fail to signal to other scientists and, perhaps more importantly, the public that these norms are being upheld. They could do so as they generate, certify, and react to each other’s findings: for example, by promoting the use and value of evidence, transparent reporting, self-correction, replication, a culture of critique, and controls for bias. A number of approaches for authors and journals would lead to more effective signals of trustworthiness at the article level. These include article badging, checklists, a more extensive withdrawal ontology, identity verification, better forward linking, and greater transparency. less

The use of NLP in the realm of financial technology is broad and complex, with applications ranging from sentiment analysis and named entity recognition to question answering. Large Language Models (LLMs) have been shown to be effective on a variety of tasks; however, no LLM specialized for the financial domain has been reported in literature. In this work, we present BloombergGPT, a 50 billion parameter language model that is trained on a wide range of financial data. We construct a 363 billion token dataset based on Bloomberg's extensive data sources, perhaps the largest domain-specific dataset yet, augmented with 345 billion tokens from general purpose datasets. We validate BloombergGPT on standard LLM benchmarks, open financial benchmarks, and a suite of internal benchmarks that most accurately reflect our intended usage. Our mixed dataset training leads to a model that outperforms existing models on financial tasks by significant margins without sacrificing performance on general LLM benchmarks. Additionally, we explain our modeling choices, training process, and evaluation methodology. As a next step, we plan to release training logs (Chronicles) detailing our experience in training BloombergGPT. less

A presentation at McKinsey round table on quantum technologies. less

Proportional-integral-derivative (PID) control underlies more than $97\%$ of automated industrial processes. Controlling these processes effectively with respect to some specified set of performance goals requires finding an optimal set of PID parameters to moderate the PID loop. Tuning these parameters is a long and exhaustive process. A method (patent pending) based on deep reinforcement learning is presented that learns a relationship between generic system properties (e.g. resonance frequency), a multi-objective performance goal and optimal PID parameter values. Performance is demonstrated in the context of a real optical switching product of the foremost manufacturer of such devices globally. Switching is handled by piezoelectric actuators where switching time and optical loss are derived from the speed and stability of actuator-control processes respectively. The method achieves a $5\times$ improvement in the number of actuators that fall within the most challenging target switching speed, $\geq 20\%$ improvement in mean switching speed at the same optical loss and $\geq 75\%$ reduction in performance inconsistency when temperature varies between 5 and 73 degrees celcius. Furthermore, once trained (which takes $\mathcal{O}(hours)$), the model generates actuator-unique PID parameters in a one-shot inference process that takes $\mathcal{O}(ms)$ in comparison to up to $\mathcal{O}(week)$ required for conventional tuning methods, therefore accomplishing these performance improvements whilst achieving up to a $10^6\times$ speed-up. After training, the method can be applied entirely offline, incurring effectively zero optimisation-overhead in production. less