Training-free video understanding methods leverage the strong image comprehension capabilities of pre-trained vision language models (VLMs) by treating videos as a sequences of static frames, thus obviating the need for costly video-specific training. However, this paradigm often suffers from severe visual redundancy and high computational overhead, especially when processing long videos. Crucially, existing keyframe selection strategies, especially those based on CLIP similarity, are prone to biases and may inadvertently overlook critical frames, resulting in suboptimal video comprehension. To address these significant challenges, we propose KTV, a novel two-stage framework for efficient and effective training-free video understanding. In the first stage, KTV performs question-agnostic keyframe selection by clustering frame-level visual features, yielding a compact, diverse, and representative subset of frames that mitigates temporal redundancy. In the second stage, KTV applies key visual token selection, pruning redundant or less informative tokens from each selected keyframe based on token importance and redundancy, which significantly reduces the number of tokens fed into the LLM. Extensive experiments on the Multiple-Choice VideoQA task demonstrate that KTV outperforms state-of-the-art training-free baselines while using significantly fewer visual tokens, e.g., only 504 tokens for a 60 min video with 10800 frames, achieving 44.8% accuracy on the MLVU-Test benchmark. In particular, KTV also exceeds several training-based approaches on certain benchmarks.

Generalized Category Discovery (GCD) is a challenging task that aims to identify both seen and novel categories in unlabeled data. We argue that a clear margin between seen and novel class representations is essential for accurate recognition. However, existing methods often ignore this margin, mapping representations to prototypes without enforcing separation between seen and novel classes. This leads to a bias where seen samples are misclassified as novel. To address this issue, we propose DebiasGCD, a debiasing framework that enhances prototype separation through margin-aware learning. Unlike prior work that relies on static prototype learning and overlooks fine-grained representations, our method introduces Dynamic Prototype Debiasing (DPD) and Spatial-Aware Representation Distillation (SARD) to mitigate this bias. First, DPD dynamically enforces inter-prototype margins, improving class-specific feature learning and prototype discrimination. Meanwhile, SARD promotes local representation of spatial learning, supporting DPD to capture subtle details that further refine class-specific features. By synergizing these components, DebiasGCD significantly improves prototype discriminability, generating more reliable predictions for seen classes. Extensive experiments demonstrate that our approach effectively mitigates pseudo-labeling bias across datasets, especially on fine-grained ones, achieving +8.3% and +9.6% improvements on the ‘All’ classes in CUB and Stanford Cars, respectively.

Large Language Models (LLMs) have demonstrated powerful performance in sequential recommendation due to their robust language modeling and comprehension capabilities. In such paradigms, the item texts of interaction sequences are formulated as sentences and LLMs are utilized to learn language representations or directly generate target item texts by incorporating instructions. Despite their promise, these methods solely focus on modeling the mapping from sequential texts to target items, neglecting the relationship between the items in an interaction sequence. This results in a failure to learn the transition patterns between items, which reflect the dynamic change in user preferences and are crucial for predicting the next item. To tackle this issue, we propose a novel framework for mapping the sequential item texts to the sequential item IDs, named ST2SI. Specifically, we first introduce multi-query input and item linear projection (ILP) to model the conditional probability distribution of items. Then, we further propose ID alignment to address misalignment between item texts and item IDs by instruction tuning. Finally, we propose efficient ILP tuning to adapt flexibly to different scenarios, requiring only training a linear layer to achieve competitive performance. Extensive experiments on six real-world datasets show our approach outperforms the best baselines by 7.33% in NDCG@10, 4.65% in Recall@10, and 8.42% in MRR.

Generative recommendation has emerged as a promising paradigm aiming at directly generating the identifiers of the target candidates. Most existing methods attempt to leverage prior knowledge embedded in Pre-trained Language Models (PLMs) to improve the recommendation performance. However, they often fail to accommodate the differences between the general linguistic knowledge of PLMs and the specific needs of recommendation systems. Moreover, they rarely consider the complementary knowledge between the multimodal information of items, which represents the multi-faceted preferences of users. To facilitate efficient recommendation knowledge transfer, we propose a novel approach called Multimodal Quantitative Language for Generative Recommendation (MQL4GRec). Our key idea is to transform items from different domains and modalities into a unified language, which can serve as a bridge for transferring recommendation knowledge. Specifically, we first introduce quantitative translators to convert the text and image content of items from various domains into a new and concise language, known as quantitative language, with all items sharing the same vocabulary. Then, we design a series of quantitative language generation tasks to enrich quantitative language with semantic information and prior knowledge. Finally, we achieve the transfer of recommendation knowledge from different domains and modalities to the recommendation task through pre-training and fine-tuning. We evaluate the effectiveness of MQL4GRec through extensive experiments and comparisons with existing methods, achieving improvements over the baseline by 11.18\%, 14.82\%, and 7.95\% on the NDCG metric across three different datasets, respectively.

In this paper, we study the online quantum state learning problem which is recently proposed by Aaronson et al. (2018). In this problem, the learning algorithm sequentially predicts quantum states based on observed measurements and losses and the goal is to minimize the regret. In the previous work, the existing algorithms may output mixed quantum states. However, in many scenarios, the prediction of a pure quantum state is required. In this paper, we first propose a Follow-the-Perturbed-Leader (FTPL) algorithm that can guarantee to predict pure quantum states. Theoretical analysis shows that our algorithm can achieve an O(√T) expected regret under some reasonable settings. In the case that the pure state prediction is not mandatory, we propose another deterministic learning algorithm which is simpler and more efficient. The algorithm is based on the online gradient descent (OGD) method and can also achieve an O(√T) regret bound. The main technical contribution of this result is an algorithm of projecting an arbitrary Hermitian matrix onto the set of density matrices with respect to the Frobenius norm. We think this subroutine is of independent interest and can be widely used in many other problems in the quantum computing area. In addition to the theoretical analysis, we evaluate the algorithms with a series of simulation experiments. The experimental results show that our FTPL method and OGD method outperform the existing RFTL approach proposed by Aaronson et al. (2018) in almost all settings. In the implementation of the RFTL approach, we give a closed-form solution to the algorithm. This provides an efficient, accurate, and completely executable solution to the RFTL method.

Evolution Strategies (ES), a class of black-box optimization algorithms, has recently been demonstrated to be a viable alternative to popular MDP-based RL techniques such as Qlearning and Policy Gradients. ES achieves fairly good performance on challenging reinforcement learning problems and is easier to scale in a distributed setting. However, standard ES algorithms perform one gradient update per data sample, which is not very efficient. In this paper, with the purpose of more efficient using of sampled data, we propose a novel iterative procedure that optimizes a surrogate objective function, enabling to reuse data sample for multiple epochs of updates. We prove monotonic improvement guarantee for such procedure. By making several approximations to the theoretically-justified procedure, we further develop a practical algorithm called Trust Region Evolution Strategies (TRES). Our experiments demonstrate the effectiveness of TRES on a range of popular MuJoCo locomotion tasks in the OpenAI Gym, achieving better performance than ES algorithm.

Dynamic programming is a powerful method for solving combinatorial optimization problems. However, it does not always work well, particularly for some NP-hard problems having extremely large state spaces. In this paper, we propose an approach to boost the capability of dynamic programming with neural networks. First, we replace the conventional tabular method with neural networks of polynomial sizes to approximately represent dynamic programming functions. And then we design an iterative algorithm to train the neural network with data generated from a solution reconstruction process. Our method combines the approximating ability and flexibility of neural networks and the advantage of dynamic programming in utilizing intrinsic properties of a problem. This approach can significantly reduce the space complexity and it is flexible in balancing space, running time, and accuracy. We apply the method to the Travelling Salesman Problem (TSP). The experimental results show that our approach can solve larger problems that are intractable for conventional dynamic programming and the performances are near optimal, outperforming the well-known approximation algorithms.