Self-play constitutes a fundamental paradigm for autonomous skill acquisition, whereby agents iteratively enhance their capabilities through self-directed environmental exploration (Silver et al., 2018). Conventional self-play frameworks exploit agent symmetry within zero-sum competitive settings (Balduzzi et al., 2019), yet this approach proves inadequate for open-ended learning scenarios characterized by inherent asymmetry. Human pedagogical systems exemplify asymmetric instructional frameworks wherein educators systematically construct challenges calibrated to individual learners' developmental trajectories (Bobbitt, 1918; Bengio et al., 2009). The principal challenge resides in operationalizing these asymmetric, adaptive pedagogical mechanisms within artificial systems capable of autonomously synthesizing appropriate curricula without predetermined task hierarchies. Here we present Heterogeneous Adversarial Play (HAP), an adversarial Automatic Curriculum Learning (ACL) framework that formalizes teacher-student interactions as a minimax optimization wherein task-generating instructor and problem-solving learner co-evolve through adversarial dynamics. In contrast to prevailing ACL methodologies that employ static curricula or unidirectional task selection mechanisms, HAP establishes a bidirectional feedback system wherein instructors continuously recalibrate task complexity in response to real-time learner performance metrics. Experimental validation across multi-task learning domains demonstrates that our framework achieves performance parity with state-of-the-art (SOTA) baselines while generating curricula that enhance learning efficacy in both artificial agents and human subjects.

Neural Information Processing Systems http://nips.cc/

We thank the reviewers for their valuable comments. We are glad that reviewers noted our paper as novel (R1: "idea is "Decouple the effect of capacity increase and curriculum learning": We would like to We will also move related works section as suggested. We agree that this issue is important in the field of curriculum learning. "It could be interesting to show results on the large W ebVision Benchmark": "W ould proposed curriculum change robustness to adversarial attacks": On average, our method requires 20 % fewer epochs. ImageNet, we conducted new experiments on WebVision dataset (2.3 million training images) and obtain significant Please see the first table above.

Previousworksignore the distribution of sample difficulty and we argue that this potentially degrade their agent performance.

We thank the reviewers for their feedback. Our'formulation is generic and task-agnostic and therefore has the potential'The model simplifies existing work' ( R1) and'has been applied to many loss functions and tasks without any change'The experiments cover different tasks and benchmark datasets' ( R3). 'It is misleading to claim that the paper is the first work using task-agnostic weights that do not require iterative W e do not make such a claim . We believe a simple and easy-to-use idea has potential for great impact. We review (in Section 2.1 and Section 1 from the supplementary) We therefore propose in Section 2.2 the Section 2.3); (2) handle both positive-and negative-valued losses (which justifies the squared regularizer log term'Does not brings notably new criteria in determining the sample weights' (R3.3). 'SuperLoss does not show an advantage on clean data' (R3.4).

Infants possess a remarkable ability to rapidly learn and process visual inputs. As an infant's mobility increases, so does the variety and dynamics of their visual inputs. Is this change in the properties of the visual inputs beneficial or even critical for the proper development of the visual system? To address this question, we used video recordings from infants wearing head-mounted cameras to train a variety of self-supervised learning models. Critically, we separated the infant data by age group and evaluated the importance of training with a curriculum aligned with developmental order. We found that initiating learning with the data from the youngest age group provided the strongest learning signal and led to the best learning outcomes in terms of downstream task performance. We then showed that the benefits of the data from the youngest age group are due to the slowness and simplicity of the visual experience. The results provide strong empirical evidence for the importance of the properties of the early infant experience and developmental progression in training. More broadly, our approach and findings take a noteworthy step towards reverse engineering the learning mechanisms in newborn brains using image-computable models from artificial intelligence.