No abstract

Anthropogenic warming has led to an unprecedented year-round reduction in Arctic sea ice extent. This has far-reaching consequences for indigenous and local communities, polar ecosystems, and global climate, motivating the need for accurate seasonal sea ice forecasts. While physics-based dynamical models can successfully forecast sea ice concentration several weeks ahead, they struggle to outperform simple statistical benchmarks at longer lead times. We present a probabilistic, deep learning sea ice forecasting system, IceNet. The system has been trained on climate simulations and observational data to forecast the next 6 months of monthly-averaged sea ice concentration maps. We show that IceNet advances the range of accurate sea ice forecasts, outperforming a state-of-the-art dynamical model in seasonal forecasts of summer sea ice, particularly for extreme sea ice events. This step-change in sea ice forecasting ability brings us closer to conservation tools that mitigate risks associated with rapid sea ice loss.

The original "Seven Motifs" set forth a roadmap of essential methods for the field of scientific computing, where a motif is an algorithmic method that captures a pattern of computation and data movement. 1 We present the Nine Motifs of Simulation Intelligence, a roadmap for the development and integration of the essential algorithms necessary for a merger of scientific computing, scientific simulation, and artificial intelligence. We call this merger simulation intelligence (SI), for short. We argue the motifs of simulation intelligence are interconnected and interdependent, much like the components within the layers of an operating system. Using this metaphor, we explore the nature of each layer of the simulation intelligence "operating system" stack (SI-stack) and the motifs therein:1. Multi-physics and multi-scale modeling 2. Surrogate modeling and emulation 3. Simulation-based inference 4. Causal modeling and inference 5. Agent-based modeling 6. Probabilistic programming 7. Differentiable programming 8. Open-ended optimization Machine programmingWe believe coordinated efforts between motifs offers immense opportunity to accelerate scientific discovery, from solving inverse problems in synthetic biology and climate science, to directing nuclear energy experiments and predicting emergent behavior in socioeconomic settings. We elaborate on each layer of the SI-stack, detailing the state-of-art methods, presenting examples to highlight challenges and opportunities, and advocating for specific ways to advance the motifs and the synergies from their combinations. Advancing and integrating these technologies can enable a robust and efficient hypothesis-simulation-analysis type of scientific method, which we introduce with several use-cases for human-machine teaming and automated science.

Buying a home is one of the most important buying decisions people have to make in their life. Although it is rare that the complete transaction happens online, real estate websites play an increasingly prominent role in the overall decision-making process. For example, the online real estate company Zillow hosts a database of more than 110 million homes in the U.S. and had 196 million unique users in 2019 [7]. Besides search and recommendation capabilities, today's real estate websites offer various decision aids to support buyers as well as professional and non-professional sellers in estimating the price of a property. According to Zillow, its home valuation model Zestimate R uses statistical and machine learning models to predict the market value of a property and the prediction "is within 10% of the final sale price more than 95% of the time" [6].From a theoretical perspective, real estate evaluation models are based on hedonic pricing models introduced by Lancaster and Rosen in the 1970s and 80s [9,16]. The main idea of hedonic pricing models is that the overall price of a good can be estimated by decomposing it into its constituting characteristics, determining the value contribution of each individual characteristic, and summing these contributions up to obtain the final price. In the past, most researchers implemented hedonic real estate pricing models through linear regression models [11,8]. While linear models have the advantage of being transparent and interpretable, they typically suffer from high bias, as their assumptions, such as linearity and additivity, often do not align with reality [4].

Chemical reactions can be described as the stepwise redistribution of electrons in molecules. As such, reactions are often depicted using "arrow-pushing" diagrams which show this movement as a sequence of arrows. We propose an electron path prediction model (ELECTRO) to learn these sequences directly from raw reaction data. Instead of predicting product molecules directly from reactant molecules in one shot, learning a model of electron movement has the benefits of (a) being easy for chemists to interpret, (b) incorporating constraints of chemistry, such as balanced atom counts before and after the reaction, and (c) naturally encoding the sparsity of chemical reactions, which usually involve changes in only a small number of atoms in the reactants. We design a method to extract approximate reaction paths from any dataset of atom-mapped reaction SMILES strings. Our model achieves excellent performance on an important subset of the USPTO reaction dataset, comparing favorably to the strongest baselines. Furthermore, we show that our model recovers a basic knowledge of chemistry without being explicitly trained to do so.

The study of pattern formation has benefited from reverse-engineering gene regulatory network (GRN) structure from spatio-temporal quantitative gene expression data. Traditional approaches omit tissue morphogenesis, hence focusing on systems where the timescales of pattern formation and morphogenesis can be separated. In such systems, pattern forms as an emergent property of the underlying GRN. This is not the case in many animal patterning systems, where patterning and morphogenesis are simultaneous. To address pattern formation in these systems we need to adapt our methodologies to explicitly accommodate cell movements and tissue shape changes. In this work we present a novel framework to reverse-engineer GRNs underlying pattern formation in tissues experiencing morphogenetic changes and cell rearrangements. By combination of quantitative data from live and fixed embryos we approximate gene expression trajectories (AGETs) in single cells and use a subset to reverse-engineer candidate GRNs using a Markov Chain Monte Carlo approach. GRN fit is assessed by simulating on cell tracks (live-modelling) and comparing the output to quantitative data-sets. This framework outputs candidate GRNs that recapitulate pattern formation at the level of the tissue and the single cell. To our knowledge, this inference methodology is the first to integrate cell movements and gene expression data, making it possible to reverse-engineer GRNs patterning tissues undergoing morphogenetic changes.

Variational autoencoders (VAEs) learn representations of data by jointly training a probabilistic encoder and decoder network. Typically these models encode all features of the data into a single variable. Here we are interested in learning disentangled representations that encode distinct aspects of the data into separate variables. We propose to learn such representations using model architectures that generalise from standard VAEs, employing a general graphical model structure in the encoder and decoder. This allows us to train partially-specified models that make relatively strong assumptions about a subset of interpretable variables and rely on the flexibility of neural networks to learn representations for the remaining variables. We further define a general objective for semi-supervised learning in this model class, which can be approximated using an importance sampling procedure. We evaluate our framework's ability to learn disentangled representations, both by qualitative exploration of its generative capacity, and quantitative evaluation of its discriminative ability on a variety of models and datasets.

This is a non-peer reviewed preprint submitted to EarthArXiv. If published in a journal, a link to the final version of the manuscript will be available via the 'Peer-reviewed Publication DOI' link on the right-hand side of this webpage. Anthropogenic warming has led to an unprecedented year-round reduction in Arctic sea ice extent 1,2 . This has far-reaching consequences for indigenous and local communities, polar ecosystems, and global climate, motivating the need for accurate seasonal sea ice forecasts. While physics-based dynamical models can successfully forecast sea ice concentration several weeks ahead, they struggle to outperform simple statistical models at longer lead times 3,4 and calibrating their forecasts can be challenging 5 . We present a probabilistic, deep learning 6 sea ice forecasting system, IceNet. The system has been trained on climate simulations covering 1850-2100 and observational data from 1979-2011 to forecast the next 6 months of monthly-averaged sea ice concentration maps. IceNet advances the range of accurate sea ice forecasts, outperforming a state-of-the-art dynamical model 7 in seasonal forecasts of summer sea ice. It also demonstrates a greater ability to predict anomalous pan-Arctic sea ice extents than the models submitted to the Sea Ice Outlook programme 8 . In addition, IceNet's well-calibrated probabilistic forecasts mean it can reliably bound the ice edge between two contours. IceNet's accuracy and reliability represent a step-change in sea ice forecasting, providing a robust framework to build early-warning systems and conservation tools that mitigate risks associated with rapid sea ice loss.Near-surface air temperatures in the Arctic have increased at roughly twice the rate of the global average, a phenomenon known as 'Arctic amplification', caused by a number of positive feedbacks 1,2,9 . Rising temperatures have played a key role in reducing Arctic sea ice, with September sea ice extent now around half that of 1979 when satellite measurements of the Arctic began 10 . This downward trend will continue, even in optimistic greenhouse gas emission reduction scenarios 11 . Climate simulations project the Arctic to be ice free in the summer by 2050 12 . Other studies put this date as early as the 2030s 13 . Such unprecedented sea ice loss has profound local and regional consequences: it is the greatest threat to polar bear populations 14 ; it has increased the intensity and frequency of algal blooms that propagate toxins throughout the food web 15 ; and it poses significant challenges for Indigenous Peoples, with impacts ranging from food security 15 to loss of culture 16 .