The usage of environment sensor models for virtual testing is a promising approach to reduce the testing effort of autonomous driving. However, in order to deduce any statements regarding the performance of an autonomous driving function based on simulation, the sensor model has to be validated to determine the discrepancy between the synthetic and real sensor data. Since a certain degree of divergence can be assumed to exist, the sufficient level of fidelity must be determined, which poses a major challenge. In particular, a method for quantifying the fidelity of a sensor model does not exist and the problem of defining an appropriate metric remains. In this work, we train a neural network to distinguish real and simulated radar sensor data with the purpose of learning the latent features of real radar point clouds. Furthermore, we propose the classifier's confidence score for the `real radar point cloud' class as a metric to determine the degree of fidelity of synthetically generated radar data. The presented approach is evaluated and it can be demonstrated that the proposed deep evaluation metric outperforms conventional metrics in terms of its capability to identify characteristic differences between real and simulated radar data.

In this work, we study hallucinations in Neural Machine Translation (NMT), which lie at an extreme end on the spectrum of NMT pathologies. Firstly, we connect the phenomenon of hallucinations under source perturbation to the Long-Tail theory of Feldman (2020), and present an empirically validated hypothesis that explains hallucinations under source perturbation. Secondly, we consider hallucinations under corpus-level noise (without any source perturbation) and demonstrate that two prominent types of natural hallucinations (detached and oscillatory outputs) could be generated and explained through specific corpus-level noise patterns. Finally, we elucidate the phenomenon of hallucination amplification in popular data-generation processes such as Backtranslation and sequence-level Knowledge Distillation.

Multi-party learning provides solutions for training joint models with decentralized data under legal and practical constraints. However, traditional multi-party learning approaches are confronted with obstacles such as system heterogeneity, statistical heterogeneity, and incentive design. How to deal with these challenges and further improve the efficiency and performance of multi-party learning has become an urgent problem to be solved. In this paper, we propose a novel contrastive multi-party learning framework for knowledge refinement and sharing with an accountable incentive mechanism. Since the existing naive model parameter averaging method is contradictory to the learning paradigm of neural networks, we simulate the process of human cognition and communication, and analogy multi-party learning as a many-to-one knowledge sharing problem. The approach is capable of integrating the acquired explicit knowledge of each client in a transparent manner without privacy disclosure, and it reduces the dependence on data distribution and communication environments. The proposed scheme achieves significant improvement in model performance in a variety of scenarios, as we demonstrated through experiments on several real-world datasets.

We introduce an evaluation methodology for visual question answering (VQA) to better diagnose cases of shortcut learning. These cases happen when a model exploits spurious statistical regularities to produce correct answers but does not actually deploy the desired behavior. There is a need to identify possible shortcuts in a dataset and assess their use before deploying a model in the real world. The research community in VQA has focused exclusively on question-based shortcuts, where a model might, for example, answer "What is the color of the sky" with "blue" by relying mostly on the question-conditional training prior and give little weight to visual evidence. We go a step further and consider multimodal shortcuts that involve both questions and images. We first identify potential shortcuts in the popular VQA v2 training set by mining trivial predictive rules such as co-occurrences of words and visual elements. We then create VQA-CE, a new evaluation set made of CounterExamples i.e. questions where the mined rules lead to incorrect answers. We use this new evaluation in a large-scale study of existing models. We demonstrate that even state-of-the-art models perform poorly and that existing techniques to reduce biases are largely ineffective in this context. Our findings suggest that past work on question-based biases in VQA has only addressed one facet of a complex issue. The code for our method is available at https://github.com/cdancette/detect-shortcuts

Why 2017 Is The Year Of Artificial Intelligence.

A physicist has used the power of artificial intelligence (AI) to solve the age-old debate about whether Jaffa Cakes are biscuits or cakes. Dr. Héloïse Stevance, an astrophysicist at the University of Auckland in New Zealand, trained algorithms with nearly 100 recipes of traditional cakes and biscuits. She then ran two Jaffa Cakes recipes through the algorithms, which recognised them unambiguously as cakes'without a doubt'. Jaffa Cakes, which are made by Edinburgh-based manufacturer McVitie's, consist of a disc of orange-flavoured jelly, milk chocolate and a mysterious spongy base. But fans of the popular British snack have passionately debated whether they're biscuits or cakes due to their unique texture and appearance.

Probing neural models for the ability to perform downstream tasks using their activation patterns is often used to localize what parts of the network specialize in performing what tasks. However, little work addressed potential mediating factors in such comparisons. As a test-case mediating factor, we consider the prediction's context length, namely the length of the span whose processing is minimally required to perform the prediction. We show that not controlling for context length may lead to contradictory conclusions as to the localization patterns of the network, depending on the distribution of the probing dataset. Indeed, when probing BERT with seven tasks, we find that it is possible to get 196 different rankings between them when manipulating the distribution of context lengths in the probing dataset. We conclude by presenting best practices for conducting such comparisons in the future.

Discontinuous constituent parsers have always lagged behind continuous approaches in terms of accuracy and speed, as the presence of constituents with discontinuous yield introduces extra complexity to the task. However, a discontinuous tree can be converted into a continuous variant by reordering tokens. Based on that, we propose to reduce discontinuous parsing to a continuous problem, which can then be directly solved by any off-the-shelf continuous parser. To that end, we develop a Pointer Network capable of accurately generating the continuous token arrangement for a given input sentence and define a bijective function to recover the original order. Experiments on the main benchmarks with two continuous parsers prove that our approach is on par in accuracy with purely discontinuous state-of-the-art algorithms, but considerably faster.

This paper presents a new algorithm named spherical vector-based particle swarm optimization (SPSO) to deal with the problem of path planning for unmanned aerial vehicles (UAVs) in complicated environments subjected to multiple threats. A cost function is first formulated to convert the path planning into an optimization problem that incorporates requirements and constraints for the feasible and safe operation of the UAV. SPSO is then used to find the optimal path that minimizes the cost function by efficiently searching the configuration space of the UAV via the correspondence between the particle position and the speed, turn angle and climb/dive angle of the UAV. To evaluate the performance of SPSO, eight benchmarking scenarios have been generated from real digital elevation model maps. The results show that the proposed SPSO outperforms not only other particle swarm optimization (PSO) variants including the classic PSO, phase angle-encoded PSO and quantum-behave PSO but also other state-of-the-art metaheuristic optimization algorithms including the genetic algorithm (GA), artificial bee colony (ABC), and differential evolution (DE) in most scenarios. In addition, experiments have been conducted to demonstrate the validity of the generated paths for real UAV operations. Source code of the algorithm can be found at https://github.com/duongpm/SPSO.

Deep reinforcement learning (DRL) has achieved great successes in many simulated tasks. The sample inefficiency problem makes applying traditional DRL methods to real-world robots a great challenge. Generative Adversarial Imitation Learning (GAIL) -- a general model-free imitation learning method, allows robots to directly learn policies from expert trajectories in large environments. However, GAIL shares the limitation of other imitation learning methods that they can seldom surpass the performance of demonstrations. In this paper, to address the limit of GAIL, we propose GAN-Based Interactive Reinforcement Learning (GAIRL) from demonstration and human evaluative feedback by combining the advantages of GAIL and interactive reinforcement learning. We tested our proposed method in six physics-based control tasks, ranging from simple low-dimensional control tasks -- Cart Pole and Mountain Car, to difficult high-dimensional tasks -- Inverted Double Pendulum, Lunar Lander, Hopper and HalfCheetah. Our results suggest that with both optimal and suboptimal demonstrations, a GAIRL agent can always learn a more stable policy with optimal or close to optimal performance, while the performance of the GAIL agent is upper bounded by the performance of demonstrations or even worse than it. In addition, our results indicate the reason that GAIRL is superior over GAIL is the complementary effect of demonstrations and human evaluative feedback.