Continual learning (CL) aims to enable models to incrementally learn from a sequence of tasks without forgetting previously acquired knowledge. While most prior work focuses on closed-world settings, where all test instances are assumed from the set of learned classes, real-world applications require models to handle both CL and out-of-distribution (OOD) samples. A key insight from recent studies on deep neural networks is the phenomenon of Neural Collapse (NC), which occurs in the terminal phase of training when the loss approaches zero. Under NC, class features collapse to their means, and classifier weights align with these means, enabling effective prototype-based strategies such as nearest class mean, for both classification and OOD detection. However, in CL, catastrophic forgetting (CF) prevents the model from naturally reaching this desirable regime. In this paper, we propose a novel method called Analytic Neural Collapse (AnaNC) that analytically creates the NC properties in the feature space of a frozen pre-trained model with no training, overcoming CF. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods in continual OOD detection and learning, highlighting the effectiveness of our method in this challenging scenario.

Length generalization (LG) is a challenging problem in learning to reason. It refers to the phenomenon that when trained on reasoning problems of smaller lengths/sizes, the model struggles with problems of larger sizes or lengths. Although it has been proven that reasoning can be learned if the intermediate reasoning steps (also known as chain-of-thought (CoT)) are given in the training data, existing studies only apply to within a given length (interpolation), while LG is about extrapolation beyond the given length. This paper begins by presenting a theorem that identifies the root cause of the LG problem. It then defines a class of reasoning problems for which achieving LG with Transformers can be theoretically guaranteed, provided the CoT schemes are constructed to meet a proposed condition called $(n,r)$-consistency.

This paper studies the problem of class-incremental learning (CIL), a core setting within continual learning where a model learns a sequence of tasks, each containing a distinct set of classes. Traditional CIL methods, which do not leverage pre-trained models (PTMs), suffer from catastrophic forgetting (CF) due to the need to incrementally learn both feature representations and the classifier. The integration of PTMs into CIL has recently led to efficient approaches that treat the PTM as a fixed feature extractor combined with analytic classifiers, achieving state-of-the-art performance. However, they still face a major limitation: the inability to continually adapt feature representations to best suit the CIL tasks, leading to suboptimal performance. To address this, we propose AnaCP (Analytic Contrastive Projection), a novel method that preserves the efficiency of analytic classifiers while enabling incremental feature adaptation without gradient-based training, thereby eliminating the CF caused by gradient updates. Our experiments show that AnaCP not only outperforms existing baselines but also achieves the accuracy level of joint training, which is regarded as the upper bound of CIL.

Tunnel seismic prediction (TSP) faces challenges in accurately imaging geological anomalies due to the confined spaces of subterranean tunnels in observation, differing markedly from the open spaces encountered in surface seismic exploration. Conventional prestack depth migration approaches, such as Kirchhoff and reverse time migration, are prone to producing annular artifacts centered on the survey line, as these standard imaging techniques cannot discriminate the propagation direction of reflected waves, imaging them indistinctly onto all possible imaging points. Although previous studies utilizing the polarization attributes of seismic waves can mitigate such symmetrical artifacts to some extent, they fall short of fully suppressing them, particularly for reflectors at small azimuth angles relative to the survey line. This paper proposes an innovative solution, Whole-Space Kirchhoff Polarization Depth Migration (WSKPDM), designed to thoroughly eliminate these migration artifacts. WSKPDM utilizes polarization information extracted from multiple seismic traces to ascertain the true propagation directions of reflected waves, thereby imaging only veritable reflection points. This effectively removes symmetrical artifacts, enhancing the visualization of subsurface geological structure. Testing on synthetic seismic data from diverse fault models and real-world tunnel data shows that WSKPDM surpasses conventional techniques in imaging accuracy. Moreover, WSKPDM demonstrates robust noise immunity even with random noise and reversed polarity in seismic data. With minimal adjustments to account for the distinct polarization directions of S-wave and P-wave, WSKPDM also allows efficient S-wave imaging. This promising WSKPDM method significantly improves the clarity and reliability of TSP imaging, providing a crucial tool for safer and more efficient tunnel construction.

Strategy card game is a well-known genre that is demanding on the intelligent game-play and can be an ideal test-bench for AI. Previous work combines an end-to-end policy function and an optimistic smooth fictitious play, which shows promising performances on the strategy card game Legend of Code and Magic. In this work, we apply such algorithms to Hearthstone, a famous commercial game that is more complicated in game rules and mechanisms. We further propose several improved techniques and consequently achieve significant progress. For a machine-vs-human test we invite a Hearthstone streamer whose best rank was top 10 of the official league in China region that is estimated to be of millions of players. Our models defeat the human player in all Best-of-5 tournaments of full games (including both deck building and battle), showing a strong capability of decision making.

This paper studies the challenging continual learning (CL) setting of Class Incremental Learning (CIL). CIL learns a sequence of tasks consisting of disjoint sets of concepts or classes. At any time, a single model is built that can be applied to predict/classify test instances of any classes learned thus far without providing any task related information for each test instance. Although many techniques have been proposed for CIL, they are mostly empirical. It has been shown recently that a strong CIL system needs a strong within-task prediction (WP) and a strong out-of-distribution (OOD) detection for each task. However, it is still not known whether CIL is actually learnable. This paper shows that CIL is learnable. Based on the theory, a new CIL algorithm is also proposed. Experimental results demonstrate its effectiveness.

Recent years have witnessed the profound influence of AI technologies on computer gaming. While grandmaster-level AI robots have largely come true for complex games based on heavy back-end support, in practice many game developers crave for participant AI robots (PARs) that behave like average-level humans with inexpensive infrastructures. Unfortunately, to date there has not been a satisfactory solution that registers large-scale use. In this work, we attempt to develop practical PARs (dubbed ParliRobo) showing acceptably humanoid behaviors with well affordable infrastructures under a challenging scenario-a 3D-FPS (first-person shooter) mobile MMOG with real-time interaction requirements. Based on comprehensive real-world explorations, we eventually enable our attempt through a novel ?transform and polish" methodology. It achieves ultralight implementations of the core system components by non-intuitive yet principled approaches, and meanwhile carefully fixes the probable side effect incurred on user perceptions. Evaluation results from large-scale deployment indicate the close resemblance (96% on average) in biofidelity metrics between ParliRobo and human players; moreover, in 73% mini Turing tests ParliRobo cannot be distinguished from human players.

To obtain higher sample efficiency and superior final performance simultaneously has been one of the major challenges for deep reinforcement learning (DRL). Previous work could handle one of these challenges but typically failed to address them concurrently. In this paper, we try to tackle these two challenges simultaneously. To achieve this, we firstly decouple these challenges into two classic RL problems: data richness and exploration-exploitation trade-off. Then, we cast these two problems into the training data distribution optimization problem, namely to obtain desired training data within limited interactions, and address them concurrently via i) explicit modeling and control of the capacity and diversity of behavior policy and ii) more fine-grained and adaptive control of selective/sampling distribution of the behavior policy using a monotonic data distribution optimization. Finally, we integrate this process into Generalized Policy Iteration (GPI) and obtain a more general framework called Generalized Data Distribution Iteration (GDI). We use the GDI framework to introduce operator-based versions of well-known RL methods from DQN to Agent57. Theoretical guarantee of the superiority of GDI compared with GPI is concluded. We also demonstrate our state-of-the-art (SOTA) performance on Arcade Learning Environment (ALE), wherein our algorithm has achieved 9620.98% mean human normalized score (HNS), 1146.39% median HNS, and surpassed 22 human world records using only 200M training frames. Our performance is comparable to Agent57’s while we consume 500 times less data. We argue that there is still a long way to go before obtaining real superhuman agents in ALE.

Continual learning (CL) learns a sequence of tasks incrementally. There are two popular CL settings, class incremental learning (CIL) and task incremental learning (TIL). A major challenge of CL is catastrophic forgetting (CF). While a number of techniques are already available to effectively overcome CF for TIL, CIL remains to be highly challenging. So far, little theoretical study has been done to provide a principled guidance on how to solve the CIL problem. This paper performs such a study. It first shows that probabilistically, the CIL problem can be decomposed into two sub-problems: Within-task Prediction (WP) and Task-id Prediction (TP). It further proves that TP is correlated with out-of-distribution (OOD) detection, which connects CIL and OOD detection. The key conclusion of this study is that regardless of whether WP and TP or OOD detection are defined explicitly or implicitly by a CIL algorithm, good WP and good TP or OOD detection are necessary and sufficient for good CIL performances. Additionally, TIL is simply WP. Based on the theoretical result, new CIL methods are also designed, which outperform strong baselines in both CIL and TIL settings by a large margin.