Existing approaches built separate classifiers to detect nonsense in dialogues. In this paper, we show that without external classifiers, dialogue models can detect errors in their own messages introspectively, by calculating the likelihood of replies that are indicative of poor messages. For example, if an agent believes its partner is likely to respond "I don't understand" to a candidate message, that message may not make sense, so an alternative message should be chosen. We evaluate our approach on a dataset from the game Diplomacy, which contains long dialogues richly grounded in the game state, on which existing models make many errors. We first show that hand-crafted replies can be effective for the task of detecting nonsense in applications as complex as Diplomacy. We then design AutoReply, an algorithm to search for such discriminative replies automatically, given a small number of annotated dialogue examples. We find that AutoReply-generated replies outperform handcrafted replies and perform on par with carefully fine-tuned large supervised models. Results also show that one single reply without much computation overheads can also detect dialogue nonsense reasonably well.

Causal and attribution studies are essential for earth scientific discoveries and critical for informing climate, ecology, and water policies. However, the current generation of methods needs to keep pace with the complexity of scientific and stakeholder challenges and data availability combined with the adequacy of data-driven methods. Unless carefully informed by physics, they run the risk of conflating correlation with causation or getting overwhelmed by estimation inaccuracies. Given that natural experiments, controlled trials, interventions, and counterfactual examinations are often impractical, information-theoretic methods have been developed and are being continually refined in the earth sciences. Here we show that transfer entropy-based causal graphs, which have recently become popular in the earth sciences with high-profile discoveries, can be spurious even when augmented with statistical significance. We develop a subsample-based ensemble approach for robust causality analysis. Simulated data, and observations in climate and ecohydrology, suggest the robustness and consistency of this approach.

Monitoring plankton populations in situ is fundamental to preserve the aquatic ecosystem. Plankton microorganisms are in fact susceptible of minor environmental perturbations, that can reflect into consequent morphological and dynamical modifications. Nowadays, the availability of advanced automatic or semi-automatic acquisition systems has been allowing the production of an increasingly large amount of plankton image data. The adoption of machine learning algorithms to classify such data may be affected by the significant cost of manual annotation, due to both the huge quantity of acquired data and the numerosity of plankton species. To address these challenges, we propose an efficient unsupervised learning pipeline to provide accurate classification of plankton microorganisms. We build a set of image descriptors exploiting a two-step procedure. First, a Variational Autoencoder (VAE) is trained on features extracted by a pre-trained neural network. We then use the learnt latent space as image descriptor for clustering. We compare our method with state-of-the-art unsupervised approaches, where a set of pre-defined hand-crafted features is used for clustering of plankton images. The proposed pipeline outperforms the benchmark algorithms for all the plankton datasets included in our analysis, providing better image embedding properties.

We introduce a method for long-distance maritime route planning in polar regions, taking into account complex changing environmental conditions. The method allows the construction of optimised routes, describing the three main stages of the process: discrete modelling of the environmental conditions using a non-uniform mesh, the construction of mesh-optimal paths, and path smoothing. In order to account for different vehicle properties we construct a series of data driven functions that can be applied to the environmental mesh to determine the speed limitations and fuel requirements for a given vessel and mesh cell, representing these quantities graphically and geospatially. In describing our results, we demonstrate an example use case for route planning for the polar research ship the RRS Sir David Attenborough (SDA), accounting for ice-performance characteristics and validating the spatial-temporal route construction in the region of the Weddell Sea, Antarctica. We demonstrate the versatility of this route construction method by demonstrating that routes change depending on the seasonal sea ice variability, differences in the route-planning objective functions used, and the presence of other environmental conditions such as currents. To demonstrate the generality of our approach, we present examples in the Arctic Ocean and the Baltic Sea. The techniques outlined in this manuscript are generic and can therefore be applied to vessels with different characteristics. Our approach can have considerable utility beyond just a single vessel planning procedure, and we outline how this workflow is applicable to a wider community, e.g. commercial and passenger shipping.

In this week's geeked-out edition of the Good News Roundup, Ukraine's jaw-dropping battlefield victories with HIMARS are documented using OSINT, South Africa implements AI technology to track dangerous locust swarms, biologists and naturalists overwhelmingly agree that gay sex is normal throughout the animal kingdom, and BirdNet proves reliable at crowdsourcing the task of identifying wild birds by their songs. In wholesome news for sci fi/space fantasy fans everywhere, Ukraine's president Zelensky continues attending technology trade shows through holograms in which he promises that Ukraine will defeat the Empire. Ukrainians are also using 3d imaging technology to preserve the cultural heritage of their country from looters and bombs, storing their data in a digital archive that will support restoration work when the invaders have been defeated. And in good news for new Ukrainian parents, the non-profit Embrace Global is making headlines for using innovative technology to provide incubators for babies in Ukraine at a tiny fraction of their usual cost. You can see their TED talk by entrepreneur Jane Chen here.

An amphibious, 'shape-shifting' drone has been created that's worthy of its own James Bond film. The'dual robot' drone, called MEDUSA (Multi-Environment Dual robot for Underwater Sample Acquisition), is able to fly through the air and land on water in order to quickly collect samples for scientific studies. It has a pod tethered to it that can be deployed underwater remotely at hard-to-reach aquatic environments. Engineers at Imperial College London use the drone to measure lake water for signs of microorganisms and algal blooms, which can pose hazards to human health. In the future, it could be used to monitor climate clues like temperature changes in Arctic seas.

The trustworthiness of neural networks is often challenged because they lack the ability to express uncertainty and explain their skill. This can be problematic given the increasing use of neural networks in high stakes decision-making such as in climate change applications. We address both issues by successfully implementing a Bayesian Neural Network (BNN), where parameters are distributions rather than deterministic, and applying novel implementations of explainable AI (XAI) techniques. The uncertainty analysis from the BNN provides a comprehensive overview of the prediction more suited to practitioners' needs than predictions from a classical neural network. Using a BNN means we can calculate the entropy (i.e. uncertainty) of the predictions and determine if the probability of an outcome is statistically significant. To enhance trustworthiness, we also spatially apply the two XAI techniques of Layer-wise Relevance Propagation (LRP) and SHapley Additive exPlanation (SHAP) values. These XAI methods reveal the extent to which the BNN is suitable and/or trustworthy. Using two techniques gives a more holistic view of BNN skill and its uncertainty, as LRP considers neural network parameters, whereas SHAP considers changes to outputs. We verify these techniques using comparison with intuition from physical theory. The differences in explanation identify potential areas where new physical theory guided studies are needed.

Modern neural language models widely used in tasks across NLP risk memorizing sensitive information from their training data. As models continue to scale up in parameters, training data, and compute, understanding memorization in language models is both important from a learning-theoretical point of view, and is practically crucial in real world applications. An open question in previous studies of memorization in language models is how to filter out "common" memorization. In fact, most memorization criteria strongly correlate with the number of occurrences in the training set, capturing "common" memorization such as familiar phrases, public knowledge or templated texts. In this paper, we provide a principled perspective inspired by a taxonomy of human memory in Psychology. From this perspective, we formulate a notion of counterfactual memorization, which characterizes how a model's predictions change if a particular document is omitted during training. We identify and study counterfactually-memorized training examples in standard text datasets. We further estimate the influence of each training example on the validation set and on generated texts, and show that this can provide direct evidence of the source of memorization at test time.

For generations, the inhabitants of the Arctic have counted on seasonal sea ice, which grows and retreats during the year. Polar bears and marine mammals rely on it as a hunting spot and a place to rest; Indigenous people fish from openings in the ice known as polynyas, and use well-known routes across the ice to travel from place to place. But the Arctic air and water has warmed three times faster than the rest of the planet since 1971, according to a May 2021 report by the Arctic Council, and this warming is causing the ice to expand and contract in unpredictable ways. Some scientists and research firms are now deploying tools powered by artificial intelligence to provide more accurate and timely forecasts of what parts of the Arctic Ocean will be covered with ice, and when. AI algorithms complement existing models that use physics to understand what's happening at the ocean's surface, a dynamic zone where cold underwater currents meet harsh winds to create floating rafts of ice.

In the next week or so, the sea ice floating atop the Arctic Ocean will shrink to its smallest size this year, as summer-warmed waters eat away at the ice's submerged edges. Record lows for sea ice levels will probably not be broken this year, scientists say. In 2020, the ice covered 3.74 million square kilometers of the Arctic at its lowest point, coming nail-bitingly close to an all-time record low. Currently, sea ice is present in just under 5 million square kilometers of Arctic waters, putting it on track to become the 10th-lowest extent of sea ice in the area since satellite record keeping began in 1979. It's an unexpected finish considering that in early summer, sea ice hit a record low for that time of year. The surprise comes in part because the best current statistical- and physics-based forecasting tools can closely predict sea ice extent only a few weeks in advance, but the accuracy of long-range forecasts falters.