Search

Visual counterfactual explanation (CF) methods modify image concepts, e.g, shape, to change a prediction to a predefined outcome while closely resembling the original query image. Unlike self-explainable models (SEMs) and heatmap techniques, they grant users the ability to examine hypothetical "what-if" scenarios. Previous CF methods either entail post-hoc training, limiting the balance between transparency and CF quality, or demand optimization during inference. To bridge the gap between transparent SEMs and CF methods, we introduce the GdVAE, a self-explainable model based on a conditional variational autoencoder (CVAE), featuring a Gaussian discriminant analysis (GDA) classifier and integrated CF explanations. Full transparency is achieved through a generative classifier that leverages class-specific prototypes for the downstream task and a closed-form solution for CFs in the latent space. The consistency of CFs is improved by regularizing the latent space with the explainer function. Extensive comparisons with existing approaches affirm the effectiveness of our method in producing high-quality CF explanations while preserving transparency. Code and models are public., Comment: Accepted paper at the ECCV 2024

Refractory inclusions in chondritic meteorites, namely amoeboid olivine aggregates (AOAs) and Ca-Al-rich inclusions (CAIs), are among the first solids to have formed in the solar system. The isotopic composition of CAIs is distinct from bulk meteorites, which either results from extreme processing of presolar carriers in the CAI-forming region, or reflects an inherited heterogeneity from the Sun's parental molecular cloud. Amoeboid olivine aggregates are less refractory than CAIs and provide a record of how the isotopic composition of solid material in the disk may have changed in time and space. However, the isotopic composition of AOAs and how this composition relates to that of CAIs and later-formed solids is unknown. Here, using new O, Ti, and Cr isotopic data for eight AOAs from the Allende CV3 chondrite, we show that CAIs and AOAs share a common isotopic composition, indicating a close genetic link and formation from the same isotopic reservoir. Because AOAs are less refractory than CAIs, this observation is difficult to reconcile with a thermal processing origin of the isotope anomalies. Instead, the common isotopic composition of CAIs and AOAs is readily accounted for in a model in which the isotopic composition of infalling material from the Sun's parental molecular cloud changed over time. In this model, CAIs and AOAs record the isotopic composition of the early infall, while later-formed solids contain a larger fraction of the later, isotopically distinct infall. This model implies that CAIs and AOAs record the isotopic composition of the Sun and suggests that the nucleosynthetic isotope heterogeneity of the solar system is predominantly produced by mixing of solar nebula condensates, which acquired their distinct isotopic compositions as a result of time-varied infall from the protosolar cloud., Comment: Published gold open access in Earth and Planetary Science Letters

Nucleosynthetic isotope anomalies in meteorites allow distinguishing between the non-carbonaceous (NC) and carbonaceous (CC) meteorite reservoirs and show that correlated isotope anomalies exist in both reservoirs. It is debated, however, whether these anomalies reflect thermal processing of presolar dust in the disk or are primordial heterogeneities inherited from the Solar System's parental molecular cloud. Here, using new high-precision 84Sr isotope data, we show that NC meteorites, Mars, and the Earth and Moon are characterized by the same 84Sr isotopic composition. This 84Sr homogeneity of the inner Solar System contrasts with the well-resolved and correlated isotope anomalies among NC meteorites observed for other elements, and most likely reflects correlated s- and (r-, p-)-process heterogeneities leading to 84Sr excess and deficits of similar magnitude which cancel each other. For the same reason there is no clearly resolved 84Sr difference between NC and CC meteorites, because in some carbonaceous chondrites the characteristic 84Sr excess of the CC reservoir is counterbalanced by an 84Sr deficit resulting from s-process variations. Nevertheless, most carbonaceous chondrites exhibit 84Sr excesses, which reflect admixture of refractory inclusions and more pronounced s-process heterogeneities in these samples. Together, the correlated variations of s-, (r-, p-)-process nuclides revealed by the 84Sr data of this study refute an origin of these isotope anomalies solely by processing of presolar dust grains, but points to primordial mixing of isotopically distinct dust reservoirs as the dominant process producing the isotopic heterogeneity of the Solar System., Comment: 18 pages 4 Figures

Carbonaceous chondrites are some of the most primitive meteorites and derive from planetesimals that formed a few million years after the beginning of the solar system. Here, using new and previously published Cr, Ti, and Te isotopic data, we show that carbonaceous chondrites exhibit correlated isotopic variations that can be accounted for by mixing among three major constituents having distinct isotopic compositions, namely refractory inclusions, chondrules, and CI chondrite-like matrix. The abundances of refractory inclusions and chondrules are coupled and systematically decrease with increasing amount of matrix. We propose that these correlated abundance variations reflect trapping of chondrule precursors, including refractory inclusions, in a pressure maximum in the disk, which is likely related to the water ice line and the ultimate formation location of Jupiter. The variable abundance of refractory inclusions/chondrules relative to matrix is the result of their distinct aerodynamical properties resulting in differential delivery rates and their preferential incorporation into chondrite parent bodies during the streaming instability, consistent with the early formation of matrix-poor and the later accretion of matrix-rich carbonaceous chondrites. Our results suggest that chondrules formed locally from isotopically heterogeneous dust aggregates which themselves derive from a wide area of the disk, implying that dust enrichment in a pressure trap was an important step to facilitate the accretion of carbonaceous chondrite parent bodies or, more generally, planetesimals in the outer solar system., Comment: 12 pages, 4 figures, 1 table. Accepted for publication in ApJL

Reliable spatial uncertainty evaluation of object detection models is of special interest and has been subject of recent work. In this work, we review the existing definitions for uncertainty calibration of probabilistic regression tasks. We inspect the calibration properties of common detection networks and extend state-of-the-art recalibration methods. Our methods use a Gaussian process (GP) recalibration scheme that yields parametric distributions as output (e.g. Gaussian or Cauchy). The usage of GP recalibration allows for a local (conditional) uncertainty calibration by capturing dependencies between neighboring samples. The use of parametric distributions such as as Gaussian allows for a simplified adaption of calibration in subsequent processes, e.g., for Kalman filtering in the scope of object tracking. In addition, we use the GP recalibration scheme to perform covariance estimation which allows for post-hoc introduction of local correlations between the output quantities, e.g., position, width, or height in object detection. To measure the joint calibration of multivariate and possibly correlated data, we introduce the quantile calibration error which is based on the Mahalanobis distance between the predicted distribution and the ground truth to determine whether the ground truth is within a predicted quantile. Our experiments show that common detection models overestimate the spatial uncertainty in comparison to the observed error. We show that the simple Isotonic Regression recalibration method is sufficient to achieve a good uncertainty quantification in terms of calibrated quantiles. In contrast, if normal distributions are required for subsequent processes, our GP-Normal recalibration method yields the best results. Finally, we show that our covariance estimation method is able to achieve best calibration results for joint multivariate calibration., Comment: Accepted to SAIAD 2022 Workshop at ECCV. Source code at https://github.com/EFS-OpenSource/calibration-framework

Calibrated confidence estimates obtained from neural networks are crucial, particularly for safety-critical applications such as autonomous driving or medical image diagnosis. However, although the task of confidence calibration has been investigated on classification problems, thorough investigations on object detection and segmentation problems are still missing. Therefore, we focus on the investigation of confidence calibration for object detection and segmentation models in this chapter. We introduce the concept of multivariate confidence calibration that is an extension of well-known calibration methods to the task of object detection and segmentation. This allows for an extended confidence calibration that is also aware of additional features such as bounding box/pixel position, shape information, etc. Furthermore, we extend the expected calibration error (ECE) to measure miscalibration of object detection and segmentation models. We examine several network architectures on MS COCO as well as on Cityscapes and show that especially object detection as well as instance segmentation models are intrinsically miscalibrated given the introduced definition of calibration. Using our proposed calibration methods, we have been able to improve calibration so that it also has a positive impact on the quality of segmentation masks as well., Comment: Book chapter in: Tim Fingerscheidt, Hanno Gottschalk, Sebastian Houben (eds.): "Deep Neural Networks and Data for Automated Driving", pp. 225--250, Springer Nature, Switzerland, 2022

Modern neural networks have found to be miscalibrated in terms of confidence calibration, i.e., their predicted confidence scores do not reflect the observed accuracy or precision. Recent work has introduced methods for post-hoc confidence calibration for classification as well as for object detection to address this issue. Especially in safety critical applications, it is crucial to obtain a reliable self-assessment of a model. But what if the calibration method itself is uncertain, e.g., due to an insufficient knowledge base? We introduce Bayesian confidence calibration - a framework to obtain calibrated confidence estimates in conjunction with an uncertainty of the calibration method. Commonly, Bayesian neural networks (BNN) are used to indicate a network's uncertainty about a certain prediction. BNNs are interpreted as neural networks that use distributions instead of weights for inference. We transfer this idea of using distributions to confidence calibration. For this purpose, we use stochastic variational inference to build a calibration mapping that outputs a probability distribution rather than a single calibrated estimate. Using this approach, we achieve state-of-the-art calibration performance for object detection calibration. Finally, we show that this additional type of uncertainty can be used as a sufficient criterion for covariate shift detection. All code is open source and available at https://github.com/EFS-OpenSource/calibration-framework., Comment: Published at 32nd IEEE Intelligent Vehicles Symposium (IV) 2021

Isotopic anomalies in chondrules hold important clues about the dynamics of mixing and transport processes in the solar accretion disk. These anomalies have been interpreted to indicate either disk-wide transport of chondrules or local heterogeneities of chondrule precursors. However, all previous studies relied on isotopic data for a single element (either Cr, Ti, or O), which does not allow distinguishing between source and precursor signatures as the cause of the chondrules isotope anomalies. Here we obtained the first combined O, Ti, and Cr isotope data for individual chondrules from enstatite, ordinary, and carbonaceous chondrites. We find that chondrules from non-carbonaceous (NC) chondrites have relatively homogeneous {\Delta}17O, {\epsilon}50Ti, and {\epsilon}54Cr, which are similar to the compositions of their host chondrites. By contrast, chondrules from carbonaceous chondrites (CC) have more variable compositions. Although the compositions of the analyzed CC and NC chondrules may overlap for either {\epsilon}50Ti, {\epsilon}54Cr, or {\Delta}17O, in multi-isotope space none of the CC chondrules plot in the compositional field of NC chondrites, and no NC chondrule plots within the field of CC chondrites. As such, our data reveal a fundamental isotopic difference between NC and CC chondrules, which is inconsistent with a disk-wide transport of chondrules across and between the NC and CC reservoirs. Instead, the isotopic variations among CC chondrules reflect local precursor heterogeneities, which most likely result from mixing between NC-like dust and a chemically diverse dust component that was isotopically similar to CAIs and AOAs.The same mixing processes, but on a larger, disk-wide scale, were likely responsible for establishing the distinct isotopic compositions of the NC and CC reservoirs, which represent in inner and outer disk, respectively., Comment: 1 Table 5 Figures Supplementary Materials: 1 Table, 1Figure

Understanding decisions made by neural networks is key for the deployment of intelligent systems in real world applications. However, the opaque decision making process of these systems is a disadvantage where interpretability is essential. Many feature-based explanation techniques have been introduced over the last few years in the field of machine learning to better understand decisions made by neural networks and have become an important component to verify their reasoning capabilities. However, existing methods do not allow statements to be made about the uncertainty regarding a feature's relevance for the prediction. In this paper, we introduce Monte Carlo Relevance Propagation (MCRP) for feature relevance uncertainty estimation. A simple but powerful method based on Monte Carlo estimation of the feature relevance distribution to compute feature relevance uncertainty scores that allow a deeper understanding of a neural network's perception and reasoning., Comment: 18 pages, 15 figures

On April 13th, 2019, OpenAI Five became the first AI system to defeat the world champions at an esports game. The game of Dota 2 presents novel challenges for AI systems such as long time horizons, imperfect information, and complex, continuous state-action spaces, all challenges which will become increasingly central to more capable AI systems. OpenAI Five leveraged existing reinforcement learning techniques, scaled to learn from batches of approximately 2 million frames every 2 seconds. We developed a distributed training system and tools for continual training which allowed us to train OpenAI Five for 10 months. By defeating the Dota 2 world champion (Team OG), OpenAI Five demonstrates that self-play reinforcement learning can achieve superhuman performance on a difficult task.

We demonstrate that models trained only in simulation can be used to solve a manipulation problem of unprecedented complexity on a real robot. This is made possible by two key components: a novel algorithm, which we call automatic domain randomization (ADR) and a robot platform built for machine learning. ADR automatically generates a distribution over randomized environments of ever-increasing difficulty. Control policies and vision state estimators trained with ADR exhibit vastly improved sim2real transfer. For control policies, memory-augmented models trained on an ADR-generated distribution of environments show clear signs of emergent meta-learning at test time. The combination of ADR with our custom robot platform allows us to solve a Rubik's cube with a humanoid robot hand, which involves both control and state estimation problems. Videos summarizing our results are available: https://openai.com/blog/solving-rubiks-cube/

We use reinforcement learning (RL) to learn dexterous in-hand manipulation policies which can perform vision-based object reorientation on a physical Shadow Dexterous Hand. The training is performed in a simulated environment in which we randomize many of the physical properties of the system like friction coefficients and an object's appearance. Our policies transfer to the physical robot despite being trained entirely in simulation. Our method does not rely on any human demonstrations, but many behaviors found in human manipulation emerge naturally, including finger gaiting, multi-finger coordination, and the controlled use of gravity. Our results were obtained using the same distributed RL system that was used to train OpenAI Five. We also include a video of our results: https://youtu.be/jwSbzNHGflM, Comment: Making OpenAI the first author. We wish this paper to be cited as "Learning Dexterous In-Hand Manipulation" by OpenAI et al. We are replicating the approach from the physics community: arXiv:1812.06489

The purpose of this technical report is two-fold. First of all, it introduces a suite of challenging continuous control tasks (integrated with OpenAI Gym) based on currently existing robotics hardware. The tasks include pushing, sliding and pick & place with a Fetch robotic arm as well as in-hand object manipulation with a Shadow Dexterous Hand. All tasks have sparse binary rewards and follow a Multi-Goal Reinforcement Learning (RL) framework in which an agent is told what to do using an additional input. The second part of the paper presents a set of concrete research ideas for improving RL algorithms, most of which are related to Multi-Goal RL and Hindsight Experience Replay.

Deep learning-based robotic grasping has made significant progress thanks to algorithmic improvements and increased data availability. However, state-of-the-art models are often trained on as few as hundreds or thousands of unique object instances, and as a result generalization can be a challenge. In this work, we explore a novel data generation pipeline for training a deep neural network to perform grasp planning that applies the idea of domain randomization to object synthesis. We generate millions of unique, unrealistic procedurally generated objects, and train a deep neural network to perform grasp planning on these objects. Since the distribution of successful grasps for a given object can be highly multimodal, we propose an autoregressive grasp planning model that maps sensor inputs of a scene to a probability distribution over possible grasps. This model allows us to sample grasps efficiently at test time (or avoid sampling entirely). We evaluate our model architecture and data generation pipeline in simulation and the real world. We find we can achieve a $>$90% success rate on previously unseen realistic objects at test time in simulation despite having only been trained on random objects. We also demonstrate an 80% success rate on real-world grasp attempts despite having only been trained on random simulated objects., Comment: 8 pages, 11 figures. Submitted to 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2018)

We derive the gravitational dynamics of the tensorial geometry which underlies the most general linear theory of electrodynamics that features weak birefringence in vacuo. This derivation is performed by way of gravitational closure, which is a mechanism that employs the causal structure of any canonically quantizable matter dynamics on some tensorial spacetime geometry in order to derive canonical dynamics for the latter. The resulting eleven-parameter family of weak gravitational field equations allows to predict where vacuum birefringence will occur, if there is any., Comment: 32 pages, no figures

Dealing with sparse rewards is one of the biggest challenges in Reinforcement Learning (RL). We present a novel technique called Hindsight Experience Replay which allows sample-efficient learning from rewards which are sparse and binary and therefore avoid the need for complicated reward engineering. It can be combined with an arbitrary off-policy RL algorithm and may be seen as a form of implicit curriculum. We demonstrate our approach on the task of manipulating objects with a robotic arm. In particular, we run experiments on three different tasks: pushing, sliding, and pick-and-place, in each case using only binary rewards indicating whether or not the task is completed. Our ablation studies show that Hindsight Experience Replay is a crucial ingredient which makes training possible in these challenging environments. We show that our policies trained on a physics simulation can be deployed on a physical robot and successfully complete the task.

Imitation learning has been commonly applied to solve different tasks in isolation. This usually requires either careful feature engineering, or a significant number of samples. This is far from what we desire: ideally, robots should be able to learn from very few demonstrations of any given task, and instantly generalize to new situations of the same task, without requiring task-specific engineering. In this paper, we propose a meta-learning framework for achieving such capability, which we call one-shot imitation learning. Specifically, we consider the setting where there is a very large set of tasks, and each task has many instantiations. For example, a task could be to stack all blocks on a table into a single tower, another task could be to place all blocks on a table into two-block towers, etc. In each case, different instances of the task would consist of different sets of blocks with different initial states. At training time, our algorithm is presented with pairs of demonstrations for a subset of all tasks. A neural net is trained that takes as input one demonstration and the current state (which initially is the initial state of the other demonstration of the pair), and outputs an action with the goal that the resulting sequence of states and actions matches as closely as possible with the second demonstration. At test time, a demonstration of a single instance of a new task is presented, and the neural net is expected to perform well on new instances of this new task. The use of soft attention allows the model to generalize to conditions and tasks unseen in the training data. We anticipate that by training this model on a much greater variety of tasks and settings, we will obtain a general system that can turn any demonstrations into robust policies that can accomplish an overwhelming variety of tasks. Videos available at https://bit.ly/nips2017-oneshot .

Bridging the 'reality gap' that separates simulated robotics from experiments on hardware could accelerate robotic research through improved data availability. This paper explores domain randomization, a simple technique for training models on simulated images that transfer to real images by randomizing rendering in the simulator. With enough variability in the simulator, the real world may appear to the model as just another variation. We focus on the task of object localization, which is a stepping stone to general robotic manipulation skills. We find that it is possible to train a real-world object detector that is accurate to $1.5$cm and robust to distractors and partial occlusions using only data from a simulator with non-realistic random textures. To demonstrate the capabilities of our detectors, we show they can be used to perform grasping in a cluttered environment. To our knowledge, this is the first successful transfer of a deep neural network trained only on simulated RGB images (without pre-training on real images) to the real world for the purpose of robotic control., Comment: 8 pages, 7 figures. Submitted to 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2017)

The Etherington distance duality relation, which relates the luminosity distance, the angular diameter distance and the redshift of objects, depends only upon a conservation law for light that traces back directly to the Lorentzian spacetime geometry. We show that this duality relation indeed survives transition to the most general linear electrodynamics without birefringence, which rests on a spacetime geometry constituted by a Lorentzian metric and two scalar fields, a dilaton and an axion. By computing the Poynting vector and optical scalar transport in the geometrical optics limit of this framework, we derive the modification of the light flux in the presence of a dilaton field and present improved constraints on the gradient of the dilaton field from observations of the Cosmic Microwave Background spectrum. Although this flux modification may seem applicable also to fundamental modifications of the Etherington relation, we show that the distance duality relation still holds true. Thus any deviations within this classical theory would imply non-metricities, once astrophysical sources for attenuation, such as dust, are accounted for. Moreover, using the most up-to-date measurements of the luminosity distances to Supernovae of Type Ia, and the inferred angular diameter distances from the baryon acoustic feature measurements, we perform a scale-free and nearly model-independent test of the Etherington distance duality relation between redshifts of $0.38$ and $0.61$. We find consistency with the standard distance duality relation and constrain the optical depth of light between these two redshifts to $\Delta\tau=-0.006\pm0.046$., Comment: 14 pages, 1 figure, submitted to PRD, comments welcome. A python wrapper for the likelihood code for the SNeIa JLA sample can be found here: http://www.github.com/surhudm/jla_python

Developing control policies in simulation is often more practical and safer than directly running experiments in the real world. This applies to policies obtained from planning and optimization, and even more so to policies obtained from reinforcement learning, which is often very data demanding. However, a policy that succeeds in simulation often doesn't work when deployed on a real robot. Nevertheless, often the overall gist of what the policy does in simulation remains valid in the real world. In this paper we investigate such settings, where the sequence of states traversed in simulation remains reasonable for the real world, even if the details of the controls are not, as could be the case when the key differences lie in detailed friction, contact, mass and geometry properties. During execution, at each time step our approach computes what the simulation-based control policy would do, but then, rather than executing these controls on the real robot, our approach computes what the simulation expects the resulting next state(s) will be, and then relies on a learned deep inverse dynamics model to decide which real-world action is most suitable to achieve those next states. Deep models are only as good as their training data, and we also propose an approach for data collection to (incrementally) learn the deep inverse dynamics model. Our experiments shows our approach compares favorably with various baselines that have been developed for dealing with simulation to real world model discrepancy, including output error control and Gaussian dynamics adaptation.

Efficient evaluation of multi-dimensional range queries in a main-memory database is an important, but difficult task. State-of-the-art techniques rely on optimised sequential scans or tree-based structures. For range queries with small result sets, sequential scans exhibit poor asymptotic performance. Also, as the dimensionality of the data set increases, the performance of tree-based structures degenerates due to the curse of dimensionality. Recent literature proposed the Elf, a main-memory structure that is optimised for the case of such multi-dimensional low-selectivity queries. The Elf outperforms other state-of-the-art methods in manually tuned scenarios. However, choosing an optimal parameter configuration for the Elf is vital, since for poor configurations, the search performance degrades rapidly. Consequently, further knowledge about the behaviour of the Elf in different configurations is required to achieve robust performance. In this thesis, we therefore propose a numerical cost model for the Elf. Like all main-memory index structures, the Elf response time is not dominated by disk accesses, refusing a straightforward analysis. Our model predicts the size and shape of the Elf region that is examined during search. We propose that the response time of a search is linear to the size of this region. Furthermore, we study the impact of skewed data distributions and correlations on the shape of the Elf. We find that they lead to behaviour that is accurately describable through simple reductions in attribute cardinality. Our experimental results indicate that for data sets of up to 15 dimensions, our cost model predicts the size of the examined Elf region with relative errors below 5%. Furthermore, we find that the size of the Elf region examined during search predicts the response time with an accuracy of 80%., Comment: Bachelor's Thesis, 84 pages. Correspondence to mail@jonasschneider.com

OpenAI Gym is a toolkit for reinforcement learning research. It includes a growing collection of benchmark problems that expose a common interface, and a website where people can share their results and compare the performance of algorithms. This whitepaper discusses the components of OpenAI Gym and the design decisions that went into the software.