Recent few-shot video classification (FSVC) works achieve promising performance by capturing similarity across support and query samples with different temporal alignment strategies or learning discriminative features via Transformer block within each episode. However, they ignore two important issues: a) It is difficult to capture rich intrinsic action semantics from a limited number of instances task. b) Redundant irrelevant frames in videos easily weaken the positive influence frames. To...

Opponent modeling is essential to exploit sub-optimal opponents in strategic interactions. Most previous works focus on building explicit models predict the opponents' styles or strategies, which require a large amount of data train model and lack adaptability unknown opponents. In this work, we propose novel Learning Exploit (L2E) framework for implicit opponent modeling. L2E acquires ability through few interactions with different during training neural network can quickly adapt new...

Experience replay plays a crucial role in Reinforcement Learning (RL), enabling the agent to remember and reuse experience from past. Most previous methods sample transitions using simple heuristics like uniformly sampling or prioritizing those good ones. Since humans can learn both bad experiences, more sophisticated algorithms need be developed. Inspired by potential energy physics, this work introduces artificial field into develops Potentialized Replay (PotER) as new effective algorithm...

This paper proposes a novel, data-agnostic, model poisoning attack on Federated Learning (FL), by designing new adversarial graph autoencoder (GAE)-based framework. The requires no knowledge of FL training data and achieves both effectiveness undetectability. By listening to the benign local models global model, attacker extracts structural correlations among features substantiating models. then adversarially regenerates while maximizing loss, subsequently generates malicious using structure...

Multi-task reinforcement learning endeavors to accomplish a set of different tasks with single policy. To enhance data efficiency by sharing parameters across multiple tasks, common practice segments the network into distinct modules and trains routing recombine these task-specific policies. However, existing approaches employ fixed number for all neglecting that varying difficulties commonly require amounts knowledge. This work presents Dynamic Depth Routing (D2R) framework, which learns...

Reinforcement learning (RL) algorithms typically require orders of magnitude more interactions than humans to learn effective policies. Research on memory in neuroscience suggests that humans' efficiency benefits from associating their experiences and reconstructing potential events. Inspired by this finding, we introduce a human brainlike structure for agents build general framework based improve the RL sampling efficiency. Since is similar reconstruction process psychology, name newly...

Owing to the unremitting efforts from a few institutes, researchers have recently made significant progress in designing superhuman artificial intelligence (AI) no-limit Texas hold'em (NLTH), primary testbed for large-scale imperfect-information game research. However, it remains challenging new study this problem since there are no standard benchmarks comparing with existing methods, which hinders further developments research area. This work presents OpenHoldem, an integrated benchmark...

Solving hard exploration tasks with sparse rewards is notoriously challenging in reinforcement learning (RL), which needs to address two key issues simultaneously: exploiting past successful experiences and exploring the unknown environment. Many prior works take expert demonstrations as learn imitate them directly. However, these are often not available practice. Recently, curiosity-driven RL methods provide intrinsic rewards, encouraging agent explore states high novelty. Nonetheless, they...

This work considers the problem of deep reinforcement learning (RL) with long time dependencies and sparse rewards, as are found in many hard exploration games. A graph-based representation is proposed to allow an agent perform self-navigation for environmental exploration. The graph not only effectively models environment structure, but also efficiently traces state changes corresponding actions. By encouraging earn a new influence-based curiosity reward game observations, whole task...

Gaussian processes (GPs), or distributions over arbitrary functions in a continuous domain, can be generalized to the multi-output case: linear model of coregionalization (LMC) is one approach. LMCs estimate and exploit correlations across multiple outputs. While estimation performed efficiently for single-output GPs, these assume stationarity, but case cross-covariance interaction not stationary. We propose Large Linear GP (LLGP), which circumvents need stationarity by inducing structure...

Opponent modeling is essential to exploit sub-optimal opponents in strategic interactions. Most previous works focus on building explicit models directly predict the opponents' styles or strategies, which require a large amount of data train model and lack adaptability unknown opponents. In this work, we propose novel Learning Exploit (L2E) framework for implicit opponent modeling. L2E acquires ability by few interactions with different during training, thus can adapt new testing quickly. We...