ABSTRACT
This paper reports the results and post-challenge analyses of ChaLearn's AutoDL challenge series, which helped sorting out a profusion of AutoML solutions for Deep Learning (DL) that had been introduced in a variety of settings, but lacked fair comparisons. All input data modalities (time series, images, videos, text, tabular) were formatted as tensors and all tasks were multi-label classification problems. Code submissions were executed on hidden tasks, with limited time and computational resources, pushing solutions that get results quickly. In this setting, DL methods dominated, though popular Neural Architecture Search (NAS) was impractical. Solutions relied on fine-tuned pre-trained networks, with architectures matching data modality. Post-challenge tests did not reveal improvements beyond the imposed time limit. While no component is particularly original or novel, a high level modular organization emerged featuring a "meta-learner", "data ingestor", "model selector", "model/learner", and "evaluator". This modularity enabled ablation studies, which revealed the importance of (off-platform) meta-learning, ensembling, and efficient data management. Experiments on heterogeneous module combinations further confirm the (local) optimality of the winning solutions. Our challenge legacy includes an ever-lasting benchmark (http://autodl.chalearn.org), the open-sourced code of the winners, and a free "AutoDL self-service."
ABSTRACT
Ongoing technological advances in the areas of computation, sensing, and mechatronics enable robotic-based systems to interact with humans in the real world. To succeed against a human in a competitive scenario, a robot must anticipate the human behavior and include it in its own planning framework. Then it can predict the next human move and counter it accordingly, thus not only achieving overall better performance but also systematically exploiting the opponent's weak spots. Pool is used as a representative scenario to derive a model-based planning and control framework where not only the physics of the environment but also a model of the opponent is considered. By representing the game of pool as a Markov decision process and incorporating a model of the human decision-making based on studies, an optimized policy is derived. This enables the robot to include the opponent's typical game style into its tactical considerations when planning a stroke. The results are validated in simulations and real-life experiments with an anthropomorphic robot playing pool against a human.
