Scaling Law for Post-training after Model Pruning

Large language models (LLMs) based on the Transformer architecture are widely employed across various domains and tasks. However, their increasing size imposes significant hardware demands, limiting practical deployment. To mitigate this, model pruning techniques have been developed to create more efficient models while maintaining high performance. Despite this, post-training after pruning is crucial for performance recovery and can be resource-intensive. This paper investigates the post-training requirements of pruned LLMs and introduces a scaling law to determine the optimal amount of post-training data. Post-training experiments with the Llama-3 and Qwen-2.5 series models, pruned using depth pruning, width pruning, and 2:4 semi-structured pruning, show that higher pruning ratios necessitate more post-training data for performance recovery, whereas larger LLMs require less. The proposed scaling law predicts a model's loss based on its parameter counts before and after pruning, as well as the post-training token counts. Furthermore, we find that the scaling law established from smaller LLMs can be reliably extrapolated to larger LLMs. This work provides valuable insights into the post-training of pruned LLMs and offers a practical scaling law for optimizing post-training data usage.

Gradient-Guided Conditional Diffusion Models for Private Image Reconstruction: Analyzing Adversarial Impacts of Differential Privacy and Denoising

We investigate the construction of gradient-guided conditional diffusion models for reconstructing private images, focusing on the adversarial interplay between differential privacy noise and the denoising capabilities of diffusion models. While current gradient-based reconstruction methods struggle with high-resolution images due to computational complexity and prior knowledge requirements, we propose two novel methods that require minimal modifications to the diffusion model's generation process and eliminate the need for prior knowledge. Our approach leverages the strong image generation capabilities of diffusion models to reconstruct private images starting from randomly generated noise, even when a small amount of differentially private noise has been added to the gradients. We also conduct a comprehensive theoretical analysis of the impact of differential privacy noise on the quality of reconstructed images, revealing the relationship among noise magnitude, the architecture of attacked models, and the attacker's reconstruction capability. Additionally, extensive experiments validate the effectiveness of our proposed methods and the accuracy of our theoretical findings, suggesting new directions for privacy risk auditing using conditional diffusion models.

VERIFIED: A Video Corpus Moment Retrieval Benchmark for Fine-Grained Video Understanding

Existing Video Corpus Moment Retrieval (VCMR) is limited to coarse-grained understanding, which hinders precise video moment localization when given fine-grained queries. In this paper, we propose a more challenging fine-grained VCMR benchmark requiring methods to localize the best-matched moment from the corpus with other partially matched candidates. To improve the dataset construction efficiency and guarantee high-quality data annotations, we propose VERIFIED, an automatic \underline{V}id\underline{E}o-text annotation pipeline to generate captions with \underline{R}el\underline{I}able \underline{FI}n\underline{E}-grained statics and \underline{D}ynamics. Specifically, we resort to large language models (LLM) and large multimodal models (LMM) with our proposed Statics and Dynamics Enhanced Captioning modules to generate diverse fine-grained captions for each video. To filter out the inaccurate annotations caused by the LLM hallucination, we propose a Fine-Granularity Aware Noise Evaluator where we fine-tune a video foundation model with disturbed hard-negatives augmented contrastive and matching losses. With VERIFIED, we construct a more challenging fine-grained VCMR benchmark containing Charades-FIG, DiDeMo-FIG, and ActivityNet-FIG which demonstrate a high level of annotation quality. We evaluate several state-of-the-art VCMR models on the proposed dataset, revealing that there is still significant scope for fine-grained video understanding in VCMR. Code and Datasets are in \href{https://github.com/hlchen23/VERIFIED}{https://github.com/hlchen23/VERIFIED}.

Generalized Sparse Additive Model with Unknown Link Function

Generalized additive models (GAM) have been successfully applied to high dimensional data analysis. However, most existing methods cannot simultaneously estimate the link function, the component functions and the variable interaction. To alleviate this problem, we propose a new sparse additive model, named generalized sparse additive model with unknown link function (GSAMUL), in which the component functions are estimated by B-spline basis and the unknown link function is estimated by a multi-layer perceptron (MLP) network. Furthermore, $\ell_{2,1}$-norm regularizer is used for variable selection. The proposed GSAMUL can realize both variable selection and hidden interaction. We integrate this estimation into a bilevel optimization problem, where the data is split into training set and validation set. In theory, we provide the guarantees about the convergence of the approximate procedure. In applications, experimental evaluations on both synthetic and real world data sets consistently validate the effectiveness of the proposed approach.

PCQPR: Proactive Conversational Question Planning with Reflection

Conversational Question Generation (CQG) enhances the interactivity of conversational question-answering systems in fields such as education, customer service, and entertainment. However, traditional CQG, focusing primarily on the immediate context, lacks the conversational foresight necessary to guide conversations toward specified conclusions. This limitation significantly restricts their ability to achieve conclusion-oriented conversational outcomes. In this work, we redefine the CQG task as Conclusion-driven Conversational Question Generation (CCQG) by focusing on proactivity, not merely reacting to the unfolding conversation but actively steering it towards a conclusion-oriented question-answer pair. To address this, we propose a novel approach, called Proactive Conversational Question Planning with self-Refining (PCQPR). Concretely, by integrating a planning algorithm inspired by Monte Carlo Tree Search (MCTS) with the analytical capabilities of large language models (LLMs), PCQPR predicts future conversation turns and continuously refines its questioning strategies. This iterative self-refining mechanism ensures the generation of contextually relevant questions strategically devised to reach a specified outcome. Our extensive evaluations demonstrate that PCQPR significantly surpasses existing CQG methods, marking a paradigm shift towards conclusion-oriented conversational question-answering systems.

Quantitative Representation of Scenario Difficulty for Autonomous Driving Based on Adversarial Policy Search

Adversarial scenario generation is crucial for autonomous driving testing because it can efficiently simulate various challenge and complex traffic conditions. However, it is difficult to control current existing methods to generate desired scenarios, such as the ones with different conflict levels. Therefore, this paper proposes a data-driven quantitative method to represent scenario difficulty. Compared with rule-based discrete scenario difficulty representation method, the proposed algorithm can achieve continuous difficulty representation. Specifically, the environment agent is introduced, and a reinforcement learning method combined with mechanism knowledge is constructed for policy search to obtain an agent with adversarial behavior. The model parameters of the environment agent at different stages in the training process are extracted to construct a policy group, and then the agents with different adversarial intensity are obtained, which are used to realize data generation in different difficulty scenarios through the simulation environment. Finally, a data-driven scenario difficulty quantitative representation model is constructed, which is used to output the environment agent policy under different difficulties. The result analysis shows that the proposed algorithm can generate reasonable and interpretable scenarios with high discrimination, and can provide quantifiable difficulty representation without any expert logic rule design. The video link is https://www.youtube.com/watch?v=GceGdqAm9Ys.

A Safe Self-evolution Algorithm for Autonomous Driving Based on Data-Driven Risk Quantification Model

Autonomous driving systems with self-evolution capabilities have the potential to independently evolve in complex and open environments, allowing to handle more unknown scenarios. However, as a result of the safety-performance trade-off mechanism of evolutionary algorithms, it is difficult to ensure safe exploration without sacrificing the improvement ability. This problem is especially prominent in dynamic traffic scenarios. Therefore, this paper proposes a safe self-evolution algorithm for autonomous driving based on data-driven risk quantification model. Specifically, a risk quantification model based on the attention mechanism is proposed by modeling the way humans perceive risks during driving, with the idea of achieving safety situation estimation of the surrounding environment through a data-driven approach. To prevent the impact of over-conservative safety guarding policies on the self-evolution capability of the algorithm, a safety-evolutionary decision-control integration algorithm with adjustable safety limits is proposed, and the proposed risk quantization model is integrated into it. Simulation and real-vehicle experiments results illustrate the effectiveness of the proposed method. The results show that the proposed algorithm can generate safe and reasonable actions in a variety of complex scenarios and guarantee safety without losing the evolutionary potential of learning-based autonomous driving systems.

A Safe and Efficient Self-evolving Algorithm for Decision-making and Control of Autonomous Driving Systems

Autonomous vehicles with a self-evolving ability are expected to cope with unknown scenarios in the real-world environment. Take advantage of trial and error mechanism, reinforcement learning is able to self evolve by learning the optimal policy, and it is particularly well suitable for solving decision-making problems. However, reinforcement learning suffers from safety issues and low learning efficiency, especially in the continuous action space. Therefore, the motivation of this paper is to address the above problem by proposing a hybrid Mechanism-Experience-Learning augmented approach. Specifically, to realize the efficient self-evolution, the driving tendency by analogy with human driving experience is proposed to reduce the search space of the autonomous driving problem, while the constrained optimization problem based on a mechanistic model is designed to ensure safety during the self-evolving process. Experimental results show that the proposed method is capable of generating safe and reasonable actions in various complex scenarios, improving the performance of the autonomous driving system. Compared to conventional reinforcement learning, the safety and efficiency of the proposed algorithm are greatly improved. The training process is collision-free, and the training time is equivalent to less than 10 minutes in the real world.

A Safety-Oriented Self-Learning Algorithm for Autonomous Driving: Evolution Starting from a Basic Model

Autonomous driving vehicles with self-learning capabilities are expected to evolve in complex environments to improve their ability to cope with different scenarios. However, most self-learning algorithms suffer from low learning efficiency and lacking safety, which limits their applications. This paper proposes a safety-oriented self-learning algorithm for autonomous driving, which focuses on how to achieve evolution from a basic model. Specifically, a basic model based on the transformer encoder is designed to extract and output policy features from a small number of demonstration trajectories. To improve the learning efficiency, a policy mixed approach is developed. The basic model provides initial values to improve exploration efficiency, and the self-learning algorithm enhances the adaptability and generalization of the model, enabling continuous improvement without external intervention. Finally, an actor approximator based on receding horizon optimization is designed considering the constraints of the environmental input to ensure safety. The proposed method is verified in a challenging mixed traffic environment with pedestrians and vehicles. Simulation and real-vehicle test results show that the proposed method can safely and efficiently learn appropriate autonomous driving behaviors. Compared reinforcement learning and behavior cloning methods, it can achieve comprehensive improvement in learning efficiency and performance under the premise of ensuring safety.

Minor DPO reject penalty to increase training robustness

Learning from human preference is a paradigm used in large-scale language model (LLM) fine-tuning step to better align pretrained LLM to human preference for downstream task. In the past it uses reinforcement learning from human feedback (RLHF) algorithm to optimize the LLM policy to align with these preferences and not to draft too far from the original model. Recently, Direct Preference Optimization (DPO) has been proposed to solve the alignment problem with a simplified RL-free method. Using preference pairs of chosen and reject data, DPO models the relative log probability as implicit reward function and optimize LLM policy using a simple binary cross entropy objective directly. DPO is quite straight forward and easy to be understood. It perform efficiently and well in most cases. In this article, we analyze the working mechanism of $\beta$ in DPO, disclose its syntax difference between RL algorithm and DPO, and understand the potential shortage brought by the DPO simplification. With these insights, we propose MinorDPO, which is better aligned to the original RL algorithm, and increase the stability of preference optimization process.