In this work, we propose an all-in-one video restoration framework that grounds degradation-aware semantic context of video frames in natural language via foundation models, offering interpretable and flexible guidance. Unlike prior art, our method assumes no degradation knowledge in train or test time and learns an approximation to the grounded knowledge such that the foundation model can be safely disentangled during inference adding no extra cost. Further, we call for standardization of benchmarks in all-in-one video restoration, and propose two benchmarks in multi-degradation setting, three-task (3D) and four-task (4D), and two time-varying composite degradation benchmarks; one of the latter being our proposed dataset with varying snow intensity, simulating how weather degradations affect videos naturally. We compare our method with prior works and report state-of-the-art performance on all benchmarks.

Diffusion Transformers (DiT) have become the de-facto model for generating high-quality visual content like videos and images. A huge bottleneck is the attention mechanism where complexity scales quadratically with resolution and video length. One logical way to lessen this burden is sparse attention, where only a subset of tokens or patches are included in the calculation. However, existing techniques fail to preserve visual quality at extremely high sparsity levels and might even incur non-negligible compute overheads. To address this concern, we propose Re-ttention, which implements very high sparse attention for visual generation models by leveraging the temporal redundancy of Diffusion Models to overcome the probabilistic normalization shift within the attention mechanism. Specifically, Re-ttention reshapes attention scores based on the prior softmax distribution history in order to preserve the visual quality of the full quadratic attention at very high sparsity levels. Experimental results on T2V/T2I models such as CogVideoX and the PixArt DiTs demonstrate that Re-ttention requires as few as 3.1% of the tokens during inference, outperforming contemporary methods like FastDiTAttn, Sparse VideoGen and MInference.

Monte Carlo Tree Search (MCTS) is an immensely popular search-based framework used for decision making. It is traditionally applied to domains where a perfect simulation model of the environment is available. We study and improve MCTS in the context where the environment model is given but imperfect. We show that the discrepancy between the model and the actual environment can lead to significant performance degradation with standard MCTS. We therefore develop Uncertainty Adapted MCTS (UA-MCTS), a more robust algorithm within the MCTS framework. We estimate the transition uncertainty in the given model, and direct the search towards more certain transitions in the state space. We modify all four MCTS phases to improve the search behavior by considering these estimates. We prove, in the corrupted bandit case, that adding uncertainty information to adapt UCB leads to tighter regret bound than standard UCB. Empirically, we evaluate UA-MCTS and its individual components on the deterministic domains from the MinAtar test suite. Our results demonstrate that UA-MCTS strongly improves MCTS in the presence of model transition errors.

Reinforcement learning (RL) algorithms are typically based on optimizing a Markov Decision Process (MDP) using the optimal Bellman equation. Recent studies have revealed that focusing the optimization of Bellman equations solely on in-sample actions tends to result in more stable optimization, especially in the presence of function approximation. Upon on these findings, in this paper, we propose an Empirical MDP Iteration (EMIT) framework. EMIT constructs a sequence of empirical MDPs using data from the growing replay memory. For each of these empirical MDPs, it learns an estimated Q-function denoted as $\widehat{Q}$. The key strength is that by restricting the Bellman update to in-sample bootstrapping, each empirical MDP converges to a unique optimal $\widehat{Q}$ function. Furthermore, gradually expanding from the empirical MDPs to the original MDP induces a monotonic policy improvement. Instead of creating entirely new algorithms, we demonstrate that EMIT can be seamlessly integrated with existing online RL algorithms, effectively acting as a regularizer for contemporary Q-learning methods. We show this by implementing EMIT for two representative RL algorithms, DQN and TD3. Experimental results on Atari and MuJoCo benchmarks show that EMIT significantly reduces estimation errors and substantially improves the performance of both algorithms.

Specialized hardware has become an indispensable component to deep neural network (DNN) acceleration. To keep up with the rapid evolution of neural networks, holistic and automated solutions for jointly optimizing both hardware (HW) architectures and software (SW) mapping have been studied. These studies face two major challenges. First, the combined HW-SW design space is vast, which hinders the finding of optimal or near-optimal designs. This issue is exacerbated for industrial cases when cycle accurate models are used for design evaluation in the joint optimization. Second, HW design is prone to overfitting to the input DNNs used in the HW-SW co-optimization. To address these issues, in this paper, we propose UNICO, an efficient Unified Co-Optimization framework with a novel Robustness metric for better HW generalization. Guided by a high-fidelity surrogate model, UNICO employs multi-objective Bayesian optimization to effectively explore the HW design space, and conducts adaptive, parallel and scalable software mapping search based on successive halving. To reduce HW overfitting, we propose a HW robustness metric by relating a HW configuration’s quality to its sensitivity in software mapping search, and quantitatively incorporate this metric to search for more robust HW design(s). We implement UNICO in open source accelerator platform, and compare it with the state-of-the-art solution HASCO. Experiments show that UNICO significantly outperforms HASCO; it finds design(s) with similar quality to HASCO up to 4 × faster, and eventually converges to better and more robust designs. Finally, we deploy UNICO for optimizing an industrial accelerator, and show that it generates enhanced HW design(s) for key real-world DNNs.

AlphaGo Zero pioneered the concept of two-head neural networks in Monte Carlo Tree Search (MCTS), where the policy output is used for prior action probability and the state-value estimate is used for leaf node evaluation. We propose a three-head neural net architecture with policy, state- and action-value outputs, which could lead to more efficient MCTS since neural leaf estimate can still be back-propagated in tree with delayed node expansion and evaluation. To effectively train the newly introduced action-value head on the same game dataset as for two-head nets, we exploit the optimal relations between parent and children nodes for data augmentation and regularization. In our experiments for the game of Hex, the action-value head learning achieves similar error as the state-value prediction of a two-head architecture. The resulting neural net models are then combined with the same Policy Value MCTS (PV-MCTS) implementation. We show that, due to more efficient use of neural net evaluations, PV-MCTS with three-head neural nets consistently performs better than the two-head ones, significantly outplaying the state-of-the-art player MoHex-CNN.

Using deep convolutional neural networks for move prediction has led to massive progress in computer Go. Like Go, Hex has a large branching factor that limits the success of shallow and selective search. We show that deep convolutional neural networks can be used to produce reliable move evaluation in the game of Hex. We begin by collecting self-play games of MoHex 2.0. We then train the neural networks by canonical maximum likelihood. The trained model was evaluated by playing against top programs Wolve and MoHex 2.0. Without any search, the resulting neural network produces similar playing strength as the highly optimized Resistance evaluation function used in Wolve. Finally, using the neural networks as prior knowledge, the reigning Monte-Carlo-tree-search-based world champion player MoHex 2.0 can be enhanced.

Proof Number search (PNS) is an effective algorithm for searching theoretical values on games with non-uniform branching factors. Focused depth-first proof number search (FDFPN) with dynamic widening was proposed for Hex where the branching factor is nearly uniform. However, FDFPN is fragile to its heuristic move ordering function. The recent advances of Convolutional Neural Networks (CNNs) have led to considerable progress in game playing. We investigate how to incorporate the strength of CNNs into solving, with application to the game of Hex. We describe FDFPN-CNN, a new focused DFPN search that uses convolutional neural networks. FDFPN-CNN integrates two CNNs trained from games played by expert players. The value approximation CNN provides reliable information for defining the widening size by estimating the value of the node to expand, while the policy CNN selects promising children nodes to the search. On 8x8 Hex, experimental results show FDFPN-CNN performs notably better than FDFPN, suggesting a promising direction for better solving Hex positions where learning from strong players is possible.