Related papers: Planck 2015 results. XII. Full Focal Plane simulat…
We present a multi-level solver for drawing constrained Gaussian realizations or finding the maximum likelihood estimate of the CMB sky, given noisy sky maps with partial sky coverage. The method converges substantially faster than existing…
We recently demonstrated the value of reconstructing SPECT data with fully 3D Monte Carlo reconstruction (F3DMC), in terms of spatial resolution and quantification. This was shown on a small cubic phantom (64 projections 10 x 10) in some…
We present a new approach to component separation in multifrequency CMB experiments by formulating the problem as that of partitioning the sky into pixel clusters such that within each pixel cluster the foregrounds have similar spectrum,…
A robust measurement of the clustering amplitude of the sub-mm population of starburst galaxies requires large-area surveys (>> 1 deg^2). The largest-format arrays subtend only 10 arcmin^2 on the sky and hence scan-mapping is a necessary…
Training large Mixture-of-Experts (MoE) models remains computationally prohibitive due to their extreme compute and memory demands. Although low-precision training promises to accelerate computation and reduce memory footprint, existing…
Mixing precisions for performance has been an ongoing trend as the modern hardware accelerators started including new, and mostly lower-precision, data formats. The advantage of using them is a great potential of performance gain and energy…
The Planck design and scanning strategy provide many levels of redundancy that can be exploited to provide tests of internal consistency. One of the most important is the comparison of the 70GHz and 100GHz channels. Based on different…
A fundamental requirement in the new generation of high resolution Cosmic Microwave Background imaging experiments is a strict control of systematic errors that must be kept at micro-K level in the final maps. Some of these errors are of…
The Planck mission will provide full-sky maps of the Cosmic Microwave Background with unprecedented angular resolution (~ 10') and sensitivity (DT / T = 10^-6). This requires cryogenically cooled, high sensitivity detectors as well as an…
Reinforcement learning (RL) for large language models (LLMs) is increasingly bottlenecked by rollout (generation), where long output sequence lengths make attention and KV-cache memory dominate end-to-end step time. FP8 offers an attractive…
Recent work has shown that 8-bit floating point (FP8) can be used for efficiently training neural networks with reduced computational cost compared to training in FP32/FP16. In this work, we investigate the use of FP8 training in a…
We experimentally evaluated the sensing-communication trade-off from the fixed-point precision MIMO equalizer using FPGA. At 7-bit, noise floor drops 100x and angular error 63%, but the communication performance saturates while the hardware…
The secondary cosmic microwave background (CMB) $B$-modes stem from the post-decoupling distortion of the polarization $E$-modes due to the gravitational lensing effect of large-scale structures. These lensing-induced $B$-modes constitute…
Planck has mapped the microwave sky in nine frequency bands between 30 and 857 GHz in temperature and seven bands between 30 and 353 GHz in polarization. In this paper we consider the problem of diffuse astrophysical component separation,…
World models provide a powerful framework for simulating environment dynamics conditioned on actions or instructions, enabling downstream tasks such as action planning or policy learning. Recent approaches leverage world models as learned…
We present the current estimate of instrumental and systematic effect uncertainties for the Planck-Low Frequency Instrument relevant to the first release of the Planck cosmological results. We give an overview of the main effects and of the…
We describe the processing of data from the Low Frequency Instrument (LFI) used in production of the Planck Early Release Compact Source Catalogue (ERCSC). In particular, we discuss the steps involved in reducing the data from telemetry…
We present a Bayesian calibration algorithm for CMB observations as implemented within the global end-to-end BeyondPlanck (BP) framework, and apply this to the Planck Low Frequency Instrument (LFI) data. Following the most recent Planck…
We release two artificial datasets, Simulated Flying Shapes and Simulated Planar Manipulator that allow to test the learning ability of video processing systems. In particular, the dataset is meant as a tool which allows to easily assess…
Phaseless Auxiliary-Field Quantum Monte Carlo (ph-AFQMC) has recently emerged as a promising method for the production of benchmark-level simulations of medium to large-sized molecules, due to its accuracy and favorable polynomial scaling…