Machine-learning system based on light could yield more powerful, efficient large language models

MIT News  August 22, 2023 Optical neural networks (ONNs) have recently emerged to process deep neural networks (DNN) tasks with high clock rates, parallelism, and low-loss data transmission. However, existing challenges for ONNs are high energy consumption due to their low electro-optic conversion efficiency, low compute density due to large device footprints and channel crosstalk, and long latency due to the lack of inline nonlinearity. An international team of researchers (USA – MIT, UCLA, industry, Germany) experimentally demonstrated a spatial-temporal-multiplexed ONN system that simultaneously overcomes all these challenges. They exploited neuron encoding with volume-manufactured micrometre-scale vertical-cavity surface-emitting laser (VCSEL) arrays […]

Researchers create a tool for accurately simulating complex systems

MIT  News May 4, 2023 Current trace-driven simulators assume that the interventions being simulated (e.g., a new algorithm) would not affect the validity of the traces. However, real-world traces are often biased by the choices algorithms make during trace collection, and hence replaying traces under an intervention may lead to incorrect results. Researchers at MIT developed a causal framework for unbiased trace-driven simulation called CausalSim. CausalSim addresses this challenge by learning a causal model of the system dynamics and latent factors capturing the underlying system conditions during trace collection. It learns these models using an initial randomized control trial (RCT) […]

Scientists improve the accuracy of weather and climate models

Phys.org  March 1, 2023 Researchers in Switzerland have developed a new modelling framework for atmospheric flow simulations for cryospheric regions called CRYOWRF. CRYOWRF couples the state-of-the-art and used atmospheric model WRF (the Weather Research and Forecasting model) with the detailed snow cover model SNOWPACK. CRYOWRF makes it feasible to simulate the dynamics of a large number of snow layers governed by grain-scale prognostic variables with online coupling to the atmosphere for multiscale simulations from the synoptic to the turbulent scales. They also introduced a scheme for blowing snow in CRYOWRF. They described the technical design goals, model capabilities and the […]

The lightness of water vapor adds heft to global climate models

Science Daily  October 24, 2022 The molar mass of water vapour is less than that of dry air, making humid air lighter than dry air at the same temperature and pressure. This effect is known as vapour buoyancy and has been considered negligibly small in large-scale climate dynamics. Using theory, reanalysis data and a hierarchy of climate models a team of researchers in the US (UC Davis, Lawrence Livermore National Laboratory, Pacific Northwest National Laboratory) showed that vapour buoyancy has a similar magnitude to thermal buoyancy in the tropical free troposphere. They also showed that vapour buoyancy makes cold air […]

New weather prediction model produces more accurate typhoon intensity forecasts

Phys.org  October 12, 2022 Researchers in China cycled and evaluated western North Pacific (WNP) typhoons of 2016 using Kalman filter (EnKF) combined with the Advanced Research Weather Research and Forecasting model (WRF). For all TC categories, the 6-h ensemble priors from the WRF/EnKF system had an appropriate amount of variance for TC tracks but had insufficient variance, they overestimated the intensity for weak storms but underestimated the intensity for strong storms. Comparison with the 5-d deterministic forecasts compared to the NCEP [US] and ECMWF [European] operational control forecasts showed that the WRF/EnKF forecasts generally had larger track errors than the […]

Better models of atmospheric ‘detergent’ can help predict climate change

Phys.org  November 1, 2021 The hydroxyl radical (OH) sets the oxidative capacity of the atmosphere and thus, profoundly affects the removal rate of pollutants and reactive greenhouse gases. OH estimates for past and future periods rely primarily on global atmospheric chemistry models. The models disagree ± 30% in mean OH and in its changes from the preindustrial to late 21st century. A simple steady-state relationship that accounts for ozone photolysis frequencies, water vapor, and the ratio of reactive nitrogen to carbon emissions explains temporal variability within most models, but not intermodal differences. A team of researchers in the US (University […]

Researchers develop a model to better understand the forces that generate tsunamis

Phys.org  June 21, 2021 Currently, there is a large gap in the predictions of tsunamis based on simplified models that consider the field complexity but do not capture the physics of the landslide as it enters the water. An international team of researchers (France, USA – UC Santa Barbara) measured the volume of a granular material and released it, causing it to collapse into a long, narrow channel filled with water. They found that while the density and diameter of the grains within a landslide had little effect on the amplitude of the wave, the total volume of the grains […]