Topo2Seq: Enhanced Topology Reasoning via Topology Sequence Learning

Extracting lane topology from perspective views (PV) is crucial for planning and control in autonomous driving. This approach extracts potential drivable trajectories for self-driving vehicles without relying on high-definition (HD) maps. However, the unordered nature and weak long-range perception of the DETR-like framework can result in misaligned segment endpoints and limited topological prediction capabilities. Inspired by the learning of contextual relationships in language models, the connectivity relations in roads can be characterized as explicit topology sequences. In this paper, we introduce Topo2Seq, a novel approach for enhancing topology reasoning via topology sequences learning. The core concept of Topo2Seq is a randomized order prompt-to-sequence learning between lane segment decoder and topology sequence decoder. The dual-decoder branches simultaneously learn the lane topology sequences extracted from the Directed Acyclic Graph (DAG) and the lane graph containing geometric information. Randomized order prompt-to-sequence learning extracts unordered key points from the lane graph predicted by the lane segment decoder, which are then fed into the prompt design of the topology sequence decoder to reconstruct an ordered and complete lane graph. In this way, the lane segment decoder learns powerful long-range perception and accurate topological reasoning from the topology sequence decoder. Notably, topology sequence decoder is only introduced during training and does not affect the inference efficiency. Experimental evaluations on the OpenLane-V2 dataset demonstrate the state-of-the-art performance of Topo2Seq in topology reasoning.

Optimizing Temperature for Language Models with Multi-Sample Inference

Multi-sample aggregation strategies, such as majority voting and best-of-N sampling, are widely used in contemporary large language models (LLMs) to enhance predictive accuracy across various tasks. A key challenge in this process is temperature selection, which significantly impacts model performance. Existing approaches either rely on a fixed default temperature or require labeled validation data for tuning, which are often scarce and difficult to obtain. This paper addresses the challenge of automatically identifying the (near)-optimal temperature for different LLMs using multi-sample aggregation strategies, without relying on task-specific validation data. We provide a comprehensive analysis of temperature's role in performance optimization, considering variations in model architectures, datasets, task types, model sizes, and predictive accuracy. Furthermore, we propose a novel entropy-based metric for automated temperature optimization, which consistently outperforms fixed-temperature baselines. Additionally, we incorporate a stochastic process model to enhance interpretability, offering deeper insights into the relationship between temperature and model performance.

Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning

Retrieval-augmented generation (RAG) is extensively utilized to incorporate external, current knowledge into large language models, thereby minimizing hallucinations. A standard RAG pipeline may comprise several components, such as query rewriting, document retrieval, document filtering, and answer generation. However, these components are typically optimized separately through supervised fine-tuning, which can lead to misalignments between the objectives of individual modules and the overarching aim of generating accurate answers in question-answering (QA) tasks. Although recent efforts have explored reinforcement learning (RL) to optimize specific RAG components, these approaches often focus on overly simplistic pipelines with only two components or do not adequately address the complex interdependencies and collaborative interactions among the modules. To overcome these challenges, we propose treating the RAG pipeline as a multi-agent cooperative task, with each component regarded as an RL agent. Specifically, we present MMOA-RAG, a Multi-Module joint Optimization Algorithm for RAG, which employs multi-agent reinforcement learning to harmonize all agents' goals towards a unified reward, such as the F1 score of the final answer. Experiments conducted on various QA datasets demonstrate that MMOA-RAG improves the overall pipeline performance and outperforms existing baselines. Furthermore, comprehensive ablation studies validate the contributions of individual components and the adaptability of MMOA-RAG across different RAG components and datasets. The code of MMOA-RAG is on https://github.com/chenyiqun/MMOA-RAG.

TFG-Flow: Training-free Guidance in Multimodal Generative Flow

Given an unconditional generative model and a predictor for a target property (e.g., a classifier), the goal of training-free guidance is to generate samples with desirable target properties without additional training. As a highly efficient technique for steering generative models toward flexible outcomes, training-free guidance has gained increasing attention in diffusion models. However, existing methods only handle data in continuous spaces, while many scientific applications involve both continuous and discrete data (referred to as multimodality). Another emerging trend is the growing use of the simple and general flow matching framework in building generative foundation models, where guided generation remains under-explored. To address this, we introduce TFG-Flow, a novel training-free guidance method for multimodal generative flow. TFG-Flow addresses the curse-of-dimensionality while maintaining the property of unbiased sampling in guiding discrete variables. We validate TFG-Flow on four molecular design tasks and show that TFG-Flow has great potential in drug design by generating molecules with desired properties.

Extended Cross-Modality United Learning for Unsupervised Visible-Infrared Person Re-identification

Unsupervised learning visible-infrared person re-identification (USL-VI-ReID) aims to learn modality-invariant features from unlabeled cross-modality datasets and reduce the inter-modality gap. However, the existing methods lack cross-modality clustering or excessively pursue cluster-level association, which makes it difficult to perform reliable modality-invariant features learning. To deal with this issue, we propose a Extended Cross-Modality United Learning (ECUL) framework, incorporating Extended Modality-Camera Clustering (EMCC) and Two-Step Memory Updating Strategy (TSMem) modules. Specifically, we design ECUL to naturally integrates intra-modality clustering, inter-modality clustering and inter-modality instance selection, establishing compact and accurate cross-modality associations while reducing the introduction of noisy labels. Moreover, EMCC captures and filters the neighborhood relationships by extending the encoding vector, which further promotes the learning of modality-invariant and camera-invariant knowledge in terms of clustering algorithm. Finally, TSMem provides accurate and generalized proxy points for contrastive learning by updating the memory in stages. Extensive experiments results on SYSU-MM01 and RegDB datasets demonstrate that the proposed ECUL shows promising performance and even outperforms certain supervised methods.

SORREL: Suboptimal-Demonstration-Guided Reinforcement Learning for Learning to Branch

Mixed Integer Linear Program (MILP) solvers are mostly built upon a Branch-and-Bound (B\&B) algorithm, where the efficiency of traditional solvers heavily depends on hand-crafted heuristics for branching. The past few years have witnessed the increasing popularity of data-driven approaches to automatically learn these heuristics. However, the success of these methods is highly dependent on the availability of high-quality demonstrations, which requires either the development of near-optimal heuristics or a time-consuming sampling process. This paper averts this challenge by proposing Suboptimal-Demonstration-Guided Reinforcement Learning (SORREL) for learning to branch. SORREL selectively learns from suboptimal demonstrations based on value estimation. It utilizes suboptimal demonstrations through both offline reinforcement learning on the demonstrations generated by suboptimal heuristics and self-imitation learning on past good experiences sampled by itself. Our experiments demonstrate its advanced performance in both branching quality and training efficiency over previous methods for various MILPs.

Adaptive$^2$: Adaptive Domain Mining for Fine-grained Domain Adaptation Modeling

Advertising systems often face the multi-domain challenge, where data distributions vary significantly across scenarios. Existing domain adaptation methods primarily focus on building domain-adaptive neural networks but often rely on hand-crafted domain information, e.g., advertising placement, which may be sub-optimal. We think that fine-grained "domain" patterns exist that are difficult to hand-craft in online advertisement. Thus, we propose Adaptive$^2$, a novel framework that first learns domains adaptively using a domain mining module by self-supervision and then employs a shared&specific network to model shared and conflicting information. As a practice, we use VQ-VAE as the domain mining module and conduct extensive experiments on public benchmarks. Results show that traditional domain adaptation methods with hand-crafted domains perform no better than single-domain models under fair FLOPS conditions, highlighting the importance of domain definition. In contrast, Adaptive$^2$ outperforms existing approaches, emphasizing the effectiveness of our method and the significance of domain mining. We also deployed Adaptive$^2$ in the live streaming scenario of Kuaishou Advertising System, demonstrating its commercial value and potential for automatic domain identification. To the best of our knowledge, Adaptive$^2$ is the first approach to automatically learn both domain identification and adaptation in online advertising, opening new research directions for this area.

Dynamic Modality-Camera Invariant Clustering for Unsupervised Visible-Infrared Person Re-identification

Unsupervised learning visible-infrared person re-identification (USL-VI-ReID) offers a more flexible and cost-effective alternative compared to supervised methods. This field has gained increasing attention due to its promising potential. Existing methods simply cluster modality-specific samples and employ strong association techniques to achieve instance-to-cluster or cluster-to-cluster cross-modality associations. However, they ignore cross-camera differences, leading to noticeable issues with excessive splitting of identities. Consequently, this undermines the accuracy and reliability of cross-modal associations. To address these issues, we propose a novel Dynamic Modality-Camera Invariant Clustering (DMIC) framework for USL-VI-ReID. Specifically, our DMIC naturally integrates Modality-Camera Invariant Expansion (MIE), Dynamic Neighborhood Clustering (DNC) and Hybrid Modality Contrastive Learning (HMCL) into a unified framework, which eliminates both the cross-modality and cross-camera discrepancies in clustering. MIE fuses inter-modal and inter-camera distance coding to bridge the gaps between modalities and cameras at the clustering level. DNC employs two dynamic search strategies to refine the network's optimization objective, transitioning from improving discriminability to enhancing cross-modal and cross-camera generalizability. Moreover, HMCL is designed to optimize instance-level and cluster-level distributions. Memories for intra-modality and inter-modality training are updated using randomly selected samples, facilitating real-time exploration of modality-invariant representations. Extensive experiments have demonstrated that our DMIC addresses the limitations present in current clustering approaches and achieve competitive performance, which significantly reduces the performance gap with supervised methods.

LDACP: Long-Delayed Ad Conversions Prediction Model for Bidding Strategy

In online advertising, once an ad campaign is deployed, the automated bidding system dynamically adjusts the bidding strategy to optimize Cost Per Action (CPA) based on the number of ad conversions. For ads with a long conversion delay, relying solely on the real-time tracked conversion number as a signal for bidding strategy can significantly overestimate the current CPA, leading to conservative bidding strategies. Therefore, it is crucial to predict the number of long-delayed conversions. Nonetheless, it is challenging to predict ad conversion numbers through traditional regression methods due to the wide range of ad conversion numbers. Previous regression works have addressed this challenge by transforming regression problems into bucket classification problems, achieving success in various scenarios. However, specific challenges arise when predicting the number of ad conversions: 1) The integer nature of ad conversion numbers exacerbates the discontinuity issue in one-hot hard labels; 2) The long-tail distribution of ad conversion numbers complicates tail data prediction. In this paper, we propose the Long-Delayed Ad Conversions Prediction model for bidding strategy (LDACP), which consists of two sub-modules. To alleviate the issue of discontinuity in one-hot hard labels, the Bucket Classification Module with label Smoothing method (BCMS) converts one-hot hard labels into non-normalized soft labels, then fits these soft labels by minimizing classification loss and regression loss. To address the challenge of predicting tail data, the Value Regression Module with Proxy labels (VRMP) uses the prediction bias of aggregated pCTCVR as proxy labels. Finally, a Mixture of Experts (MoE) structure integrates the predictions from BCMS and VRMP to obtain the final predicted ad conversion number.

Improve Vision Language Model Chain-of-thought Reasoning

Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes lack robust CoT reasoning data, relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers does not generalize well to reasoning tasks that require more detailed responses. To address this, we propose a two-fold approach. First, we distill rationales from GPT-4o model to enrich the training data and fine-tune VLMs, boosting their CoT performance. Second, we apply reinforcement learning to further calibrate reasoning quality. Specifically, we construct positive (correct) and negative (incorrect) pairs of model-generated reasoning chains, by comparing their predictions with annotated short answers. Using this pairwise data, we apply the Direct Preference Optimization algorithm to refine the model's reasoning abilities. Our experiments demonstrate significant improvements in CoT reasoning on benchmark datasets and better generalization to direct answer prediction as well. This work emphasizes the importance of incorporating detailed rationales in training and leveraging reinforcement learning to strengthen the reasoning capabilities of VLMs.