Yinan Yu

Shuai Wang, Xixi Wang, Yinan Yu

Large Language Models (LLMs) have shown remarkable capabilities across various tasks but remain prone to hallucinations in knowledge-intensive scenarios. Knowledge Base Question Answering (KBQA) mitigates this by grounding generation in Knowledge Graphs (KGs). However, most multi-hop KBQA methods rely on explicit edge traversal, making them fragile to KG incompleteness. In this paper, we proposed a novel graph-based soft prompting framework that shifts the reasoning paradigm from node-level path traversal to subgraph-level reasoning. Specifically, we employ a Graph Neural Network (GNN) to encode extracted structural subgraphs into soft prompts, enabling LLM to reason over richer structural context and identify relevant entities beyond immediate graph neighbors, thereby reducing sensitivity to missing edges. Furthermore, we introduce a two-stage paradigm that reduces computational cost while preserving good performance: a lightweight LLM first leverages the soft prompts to identify question-relevant entities and relations, followed by a more powerful LLM for evidence-aware answer generation. Experiments on four multi-hop KBQA benchmarks show that our approach achieves state-of-the-art performance on three of them, demonstrating its effectiveness. Code is available at the repository: https://github.com/Wangshuaiia/GraSP.

Shuai Wang, Yinan Yu

Large Language Models (LLMs) exhibit strong abilities in natural language understanding and generation, yet they struggle with knowledge-intensive reasoning. Structured Knowledge Graphs (KGs) provide an effective form of external knowledge representation and have been widely used to enhance performance in classical Knowledge Base Question Answering (KBQA) tasks. However, performing precise multi-hop reasoning over KGs for complex queries remains highly challenging. Most existing approaches decompose the reasoning process into a sequence of isolated steps executed through a fixed pipeline. While effective to some extent, such designs constrain reasoning flexibility and fragment the overall decision process, often leading to incoherence and the loss of critical intermediate information from earlier steps. In this paper, we introduce KG-Reasoner, an end-to-end framework that integrates multi-step reasoning into a unified "thinking" phase of a Reasoning LLM. Through Reinforcement Learning (RL), the LLM is trained to internalize the KG traversal process, enabling it to dynamically explore reasoning paths, and perform backtracking when necessary. Experiments on eight multi-hop and knowledge-intensive reasoning benchmarks demonstrate that KG-Reasoner achieves competitive or superior performance compared to the state-of-the-art methods. Codes are available at the repository: https://github.com/Wangshuaiia/KG-Reasoner.

Arsham Gholamzadeh Khoee, Yinan Yu, Robert Feldt et al.

Traditional methods for making software deployment decisions in the automotive industry typically rely on manual analysis of tabular software test data. These methods often lead to higher costs and delays in the software release cycle due to their labor-intensive nature. Large Language Models (LLMs) present a promising solution to these challenges. However, their application generally demands multiple rounds of human-driven prompt engineering, which limits their practical deployment, particularly for industrial end-users who need reliable and efficient results. In this paper, we propose GoNoGo, an LLM agent system designed to streamline automotive software deployment while meeting both functional requirements and practical industrial constraints. Unlike previous systems, GoNoGo is specifically tailored to address domain-specific and risk-sensitive systems. We evaluate GoNoGo's performance across different task difficulties using zero-shot and few-shot examples taken from industrial practice. Our results show that GoNoGo achieves a 100% success rate for tasks up to Level 2 difficulty with 3-shot examples, and maintains high performance even for more complex tasks. We find that GoNoGo effectively automates decision-making for simpler tasks, significantly reducing the need for manual intervention. In summary, GoNoGo represents an efficient and user-friendly LLM-based solution currently employed in our industrial partner's company to assist with software release decision-making, supporting more informed and timely decisions in the release process for risk-sensitive vehicle systems.

Shuai Wang, Yinan Yu

Large Language Models (LLMs) demonstrate impressive natural language capabilities but often struggle with knowledge-intensive reasoning tasks. Knowledge Base Question Answering (KBQA), which leverages structured Knowledge Graphs (KGs) exemplifies this challenge due to the need for accurate multi-hop reasoning. Existing approaches typically perform sequential reasoning steps guided by predefined pipelines, restricting flexibility and causing error cascades due to isolated reasoning at each step. To address these limitations, we propose KG-Hopper, a novel Reinforcement Learning (RL) framework that empowers compact open LLMs with the ability to perform integrated multi-hop KG reasoning within a single inference round. Rather than reasoning step-by-step, we train a Reasoning LLM that embeds the entire KG traversal and decision process into a unified ``thinking'' stage, enabling global reasoning over cross-step dependencies and dynamic path exploration with backtracking. Experimental results on eight KG reasoning benchmarks show that KG-Hopper, based on a 7B-parameter LLM, consistently outperforms larger multi-step systems (up to 70B) and achieves competitive performance with proprietary models such as GPT-3.5-Turbo and GPT-4o-mini, while remaining compact, open, and data-efficient. The code is publicly available at: https://github.com/Wangshuaiia/KG-Hopper.

Shuai Wang, Dhasarathy Parthasarathy, Robert Feldt et al.

Large language models (LLMs) have shown impressive capabilities in code generation. However, because most LLMs are trained on public domain corpora, directly applying them to real-world software development often yields low success rates, as these scenarios frequently require domain-specific knowledge. In particular, domain-specific tasks usually demand highly specialized solutions, which are often underrepresented or entirely absent in the training data of generic LLMs. To address this challenge, we propose DomAgent, an autonomous coding agent that bridges this gap by enabling LLMs to generate domain-adapted code through structured reasoning and targeted retrieval. A core component of DomAgent is DomRetriever, a novel retrieval module that emulates how humans learn domain-specific knowledge, by combining conceptual understanding with experiential examples. It dynamically integrates top-down knowledge-graph reasoning with bottom-up case-based reasoning, enabling iterative retrieval and synthesis of structured knowledge and representative cases to ensure contextual relevance and broad task coverage. DomRetriever can operate as part of DomAgent or independently with any LLM for flexible domain adaptation. We evaluate DomAgent on an open benchmark dataset in the data science domain (DS-1000) and further apply it to real-world truck software development tasks. Experimental results show that DomAgent significantly enhances domain-specific code generation, enabling small open-source models to close much of the performance gap with large proprietary LLMs in complex, real-world applications. The code is available at: https://github.com/Wangshuaiia/DomAgent.

Falk Dippel, Yinan Yu, Annika Rosengren et al.

Heart failure (HF) discharge planning depends on identifying patients at risk of deterioration or death, yet accurate prediction from routinely collected electronic health records (EHRs) remains challenging. We developed and validated sequence models for three one-year prediction tasks in a Swedish HF cohort (N = 42,820): clinical instability (a rehospitalization phenotype) and mortality after the initial in-hospital HF diagnosis, and mortality after the latest hospitalization. A modular three-component framework transforms structured EHRs into patient sequences by specifying tokenization strategies, temporal representations, and model configurations. Patient data included diagnoses, vital signs, laboratories, medications, and procedures. Autoregressive next-token prediction models consistently outperformed alternative objectives in short-context settings (<= 512 tokens). The best model (Llama) achieved AUPRCs (95% CI) of 0.555 (0.535-0.575), 0.582 (0.558-0.608), and 0.854 (0.842-0.865), with robust calibration. Ablations show Llama and Mamba variants learn efficient patient representations, with tiny configurations surpassing larger conventional baselines, indicating that model size alone does not improve performance. With limited clinical concepts or training data, Llama maintains strong performance, frequently surpassing full-data baselines. Combining clinical instability and mortality predictions defines four distinct care pathways, from standard primary care to intensive home care, supporting patient-centered decisions at discharge. These findings demonstrate accurate risk prediction from routine hospital data, provide actionable development guidance, and support post-discharge risk stratification.

Malsha Ashani Mahawatta Dona, Beatriz Cabrero-Daniel, Yinan Yu et al.

Today's Large Language Models (LLMs) have showcased exemplary capabilities, ranging from simple text generation to advanced image processing. Such models are currently being explored for in-vehicle services such as supporting perception tasks in Advanced Driver Assistance Systems (ADAS) or Autonomous Driving (AD) systems, given the LLMs' capabilities to process multi-modal data. However, LLMs often generate nonsensical or unfaithful information, known as ``hallucinations'': a notable issue that needs to be mitigated. In this paper, we systematically explore the adoption of SelfCheckGPT to spot hallucinations by three state-of-the-art LLMs (GPT-4o, LLaVA, and Llama3) when analysing visual automotive data from two sources: Waymo Open Dataset, from the US, and PREPER CITY dataset, from Sweden. Our results show that GPT-4o is better at generating faithful image captions than LLaVA, whereas the former demonstrated leniency in mislabeling non-hallucinated content as hallucinations compared to the latter. Furthermore, the analysis of the performance metrics revealed that the dataset type (Waymo or PREPER CITY) did not significantly affect the quality of the captions or the effectiveness of hallucination detection. However, the models showed better performance rates over images captured during daytime, compared to during dawn, dusk or night. Overall, the results show that SelfCheckGPT and its adaptation can be used to filter hallucinations in generated traffic-related image captions for state-of-the-art LLMs.

Malsha Ashani Mahawatta Dona, Beatriz Cabrero-Daniel, Yinan Yu et al.

Today's advanced driver assistance systems (ADAS), like adaptive cruise control or rear collision warning, are finding broader adoption across vehicle classes. Integrating such advanced, multimodal Large Language Models (LLMs) on board a vehicle, which are capable of processing text, images, audio, and other data types, may have the potential to greatly enhance passenger comfort. Yet, an LLM's hallucinations are still a major challenge to be addressed. In this paper, we systematically assessed potential hallucination detection strategies for such LLMs in the context of object detection in vision-based data on the example of pedestrian detection and localization. We evaluate three hallucination detection strategies applied to two state-of-the-art LLMs, the proprietary GPT-4V and the open LLaVA, on two datasets (Waymo/US and PREPER CITY/Sweden). Our results show that these LLMs can describe a traffic situation to an impressive level of detail but are still challenged for further analysis activities such as object localization. We evaluate and extend hallucination detection approaches when applying these LLMs to video sequences in the example of pedestrian detection. Our experiments show that, at the moment, the state-of-the-art proprietary LLM performs much better than the open LLM. Furthermore, consistency enhancement techniques based on voting, such as the Best-of-Three (BO3) method, do not effectively reduce hallucinations in LLMs that tend to exhibit high false negatives in detecting pedestrians. However, extending the hallucination detection by including information from the past helps to improve results.

Malsha Ashani Mahawatta Dona, Beatriz Cabrero-Daniel, Yinan Yu et al.

Today's advanced automotive systems are turning into intelligent Cyber-Physical Systems (CPS), bringing computational intelligence to their cyber-physical context. Such systems power advanced driver assistance systems (ADAS) that observe a vehicle's surroundings for their functionality. However, such ADAS have clear limitations in scenarios when the direct line-of-sight to surrounding objects is occluded, like in urban areas. Imagine now automated driving (AD) systems that ideally could benefit from other vehicles' field-of-view in such occluded situations to increase traffic safety if, for example, locations about pedestrians can be shared across vehicles. Current literature suggests vehicle-to-infrastructure (V2I) via roadside units (RSUs) or vehicle-to-vehicle (V2V) communication to address such issues that stream sensor or object data between vehicles. When considering the ongoing revolution in vehicle system architectures towards powerful, centralized processing units with hardware accelerators, foreseeing the onboard presence of large language models (LLMs) to improve the passengers' comfort when using voice assistants becomes a reality. We are suggesting and evaluating a concept to complement the ego vehicle's field-of-view (FOV) with another vehicle's FOV by tapping into their onboard LLM to let the machines have a dialogue about what the other vehicle ``sees''. Our results show that very recent versions of LLMs, such as GPT-4V and GPT-4o, understand a traffic situation to an impressive level of detail, and hence, they can be used even to spot traffic participants. However, better prompts are needed to improve the detection quality and future work is needed towards a standardised message interchange format between vehicles.

Arsham Gholamzadeh Khoee, Yinan Yu, Robert Feldt

Large pre-trained vision-language models like CLIP have transformed computer vision by aligning images and text in a shared feature space, enabling robust zero-shot transfer via prompting. Soft-prompting, such as Context Optimization (CoOp), effectively adapts these models for downstream recognition tasks by learning a set of context vectors. However, CoOp lacks explicit mechanisms for handling domain shifts across unseen distributions. To address this, we propose Domain-invariant Context Optimization (DiCoOp), an extension of CoOp optimized for domain generalization. By employing an adversarial training approach, DiCoOp forces the model to learn domain-invariant prompts while preserving discriminative power for classification. Experimental results show that DiCoOp consistently surpasses CoOp in domain generalization tasks across diverse visual domains.

Shuai Wang, Yinan Yu, Earl Barr et al.

Multidisciplinary Software Development (MSD) requires domain experts and developers to collaborate across incompatible formalisms and separate artifact sets. Today, even with AI coding assistants like GitHub Copilot, this process remains inefficient; individual coding tasks are semi-automated, but the workflow connecting domain knowledge to implementation is not. Developers and experts still lack a shared view, resulting in repeated coordination, clarification rounds, and error-prone handoffs. We address this gap through a graph-based workflow optimization approach that progressively replaces manual coordination with LLM-powered services, enabling incremental adoption without disrupting established practices. We evaluate our approach on \texttt{spapi}, a production in-vehicle API system at Volvo Group involving 192 endpoints, 420 properties, and 776 CAN signals across six functional domains. The automated workflow achieves 93.7\% F1 score while reducing per-API development time from approximately 5 hours to under 7 minutes, saving an estimated 979 engineering hours. In production, the system received high satisfaction from both domain experts and developers, with all participants reporting full satisfaction with communication efficiency.

Yinan Yu, Samuel Scheidegger, John-Fredrik Grönvall et al.

Single-vehicle accidents are the most common type of fatal accidents in Sweden, where a car drives off the road and runs into hazardous roadside objects. Proper installation and maintenance of protective objects, such as crash cushions and guard rails, may reduce the chance and severity of such accidents. Moreover, efficient detection and management of hazardous roadside objects also plays an important role in improving road safety. To better understand the state-of-the-art and system requirements, in this pre-study, we investigate the feasibility, implementation, limitations and scaling up of data processing pipelines for roadside object detection. In particular, we divide our investigation into three parts: the target of interest, the sensors of choice and the algorithm design. The data sources we consider in this study cover two common setups: 1) road surveying fleet - annual scans conducted by Trafikverket, the Swedish Transport Administration, and 2) consumer vehicle - data collected using a research vehicle from the laboratory of Resource for vehicle research at Chalmers (REVERE). The goal of this report is to investigate how to implement a scalable roadside object detection system towards safe road infrastructure and Sweden's Vision Zero.

Arsham Gholamzadeh Khoee, Yinan Yu, Robert Feldt

Deep neural networks (DNNs) have revolutionized artificial intelligence but often lack performance when faced with out-of-distribution (OOD) data, a common scenario due to the inevitable domain shifts in real-world applications. This limitation stems from the common assumption that training and testing data share the same distribution--an assumption frequently violated in practice. Despite their effectiveness with large amounts of data and computational power, DNNs struggle with distributional shifts and limited labeled data, leading to overfitting and poor generalization across various tasks and domains. Meta-learning presents a promising approach by employing algorithms that acquire transferable knowledge across various tasks for fast adaptation, eliminating the need to learn each task from scratch. This survey paper delves into the realm of meta-learning with a focus on its contribution to domain generalization. We first clarify the concept of meta-learning for domain generalization and introduce a novel taxonomy based on the feature extraction strategy and the classifier learning methodology, offering a granular view of methodologies. Additionally, we present a decision graph to assist readers in navigating the taxonomy based on data availability and domain shifts, enabling them to select and develop a proper model tailored to their specific problem requirements. Through an exhaustive review of existing methods and underlying theories, we map out the fundamentals of the field. Our survey provides practical insights and an informed discussion on promising research directions.

Pierre Lamart, Yinan Yu, Christian Berger

Machine Learning (ML) is continuously permeating a growing amount of application domains. Generative AI such as Large Language Models (LLMs) also sees broad adoption to process multi-modal data such as text, images, audio, and video. While the trend is to use ever-larger datasets for training, managing this data efficiently has become a significant practical challenge in the industry-double as much data is certainly not double as good. Rather the opposite is important since getting an understanding of the inherent quality and diversity of the underlying data lakes is a growing challenge for application-specific ML as well as for fine-tuning foundation models. Furthermore, information retrieval (IR) from expanding data lakes is complicated by the temporal dimension inherent in time-series data which must be considered to determine its semantic value. This study focuses on the different semantic-aware techniques to extract embeddings from mono-modal, multi-modal, and cross-modal data to enhance IR capabilities in a growing data lake. Articles were collected to summarize information about the state-of-the-art techniques focusing on applications of embedding for three different categories of data modalities.

Shuai Wang, Yinan Yu, Robert Feldt et al.

Vehicle API testing verifies whether the interactions between a vehicle's internal systems and external applications meet expectations, ensuring that users can access and control various vehicle functions and data. However, this task is inherently complex, requiring the alignment and coordination of API systems, communication protocols, and even vehicle simulation systems to develop valid test cases. In practical industrial scenarios, inconsistencies, ambiguities, and interdependencies across various documents and system specifications pose significant challenges. This paper presents a system designed for the automated testing of in-vehicle APIs. By clearly defining and segmenting the testing process, we enable Large Language Models (LLMs) to focus on specific tasks, ensuring a stable and controlled testing workflow. Experiments conducted on over 100 APIs demonstrate that our system effectively automates vehicle API testing. The results also confirm that LLMs can efficiently handle mundane tasks requiring human judgment, making them suitable for complete automation in similar industrial contexts.

Shuai Wang, Yinan Yu

Large Language Models (LLMs) excel in many natural language processing tasks but often exhibit factual inconsistencies in knowledge-intensive settings. Integrating external knowledge resources, particularly knowledge graphs (KGs), provides a transparent and updatable foundation for more reliable reasoning. Knowledge Base Question Answering (KBQA), which queries and reasons over KGs, is central to this effort, especially for complex, multi-hop queries. However, multi-hop reasoning poses two key challenges: (1)~maintaining coherent reasoning paths, and (2)~avoiding prematurely discarding critical multi-hop connections. To tackle these challenges, we introduce iQUEST, a question-guided KBQA framework that iteratively decomposes complex queries into simpler sub-questions, ensuring a structured and focused reasoning trajectory. Additionally, we integrate a Graph Neural Network (GNN) to look ahead and incorporate 2-hop neighbor information at each reasoning step. This dual approach strengthens the reasoning process, enabling the model to explore viable paths more effectively. Detailed experiments demonstrate the consistent improvement delivered by iQUEST across four benchmark datasets and four LLMs.

Josie Harrison, Alexander Hollberg, Yinan Yu

Computer vision models trained on Google Street View images can create material cadastres. However, current approaches need manually annotated datasets that are difficult to obtain and often have class imbalance. To address these challenges, this paper fine-tuned a Swin Transformer model on a synthetic dataset generated with DALL-E and compared the performance to a similar manually annotated dataset. Although manual annotation remains the gold standard, the synthetic dataset performance demonstrates a reasonable alternative. The findings will ease annotation needed to develop material cadastres, offering architects insights into opportunities for material reuse, thus contributing to the reduction of demolition waste.

Arsham Gholamzadeh Khoee, Shuai Wang, Yinan Yu et al.

Ensuring reliable software release decisions is critical in safety-critical domains such as automotive manufacturing. Release validation relies on large tabular datasets, yet manual analysis is slow, costly, and error-prone. While Large Language Models (LLMs) offer promising automation potential, they face challenges in analytical reasoning, structured data handling, and ambiguity resolution. This paper introduces GateLens, an LLM-based system for analyzing tabular data in the automotive domain. GateLens translates natural language queries into Relational Algebra (RA) expressions and generates optimized Python code. Unlike traditional multi-agent or planning-based systems that can be slow, opaque, and costly to maintain, GateLens emphasizes speed, transparency, and reliability. Experimental results show that GateLens outperforms the existing Chain-of-Thought (CoT) + Self-Consistency (SC) based system on real-world datasets, particularly in handling complex and ambiguous queries. Ablation studies confirm the essential role of the RA layer. Industrial deployment shows over 80% reduction in analysis time while maintaining high accuracy across test result interpretation, impact assessment, and release candidate evaluation. GateLens operates effectively in zero-shot settings without requiring few-shot examples or agent orchestration. This work advances deployable LLM system design by identifying key architectural features-intermediate formal representations, execution efficiency, and low configuration overhead-crucial for safety-critical industrial applications.

Yinan Yu, Samuel Scheidegger

Runtime geofencing for ground vehicles is rapidly emerging as a critical technology for enforcing Operational Design Domains (ODDs). However, existing solutions struggle to reconcile high-fidelity learning with the structural requirements of verifiable control. We address this by introducing PCARNN-DCBF, a novel pipeline integrating a Physics-encoded Control-Affine Residual Neural Network with a preview-based Discrete Control Barrier Function. Unlike generic learned models, PCARNN explicitly preserves the control-affine structure of vehicle dynamics, ensuring the linearity required for reliable optimization. This enables the DCBF to enforce polygonal keep-in constraints via a real-time Quadratic Program (QP) that handles high relative degree and mitigates actuator saturation. Experiments in CARLA across electric and combustion platforms demonstrate that this structure-preserving approach significantly outperforms analytical and unstructured neural baselines.

Yinan Yu, Falk Dippel, Christina E. Lundberg et al.

Objective: Machine learning (ML) predictive models are often developed without considering downstream value trade-offs and clinical interpretability. This paper introduces a cost-aware prediction (CAP) framework that combines cost-benefit analysis assisted by large language model (LLM) agents to communicate the trade-offs involved in applying ML predictions. Materials and Methods: We developed an ML model predicting 1-year mortality in patients with heart failure (N = 30,021, 22% mortality) to identify those eligible for home care. We then introduced clinical impact projection (CIP) curves to visualize important cost dimensions - quality of life and healthcare provider expenses, further divided into treatment and error costs, to assess the clinical consequences of predictions. Finally, we used four LLM agents to generate patient-specific descriptions. The system was evaluated by clinicians for its decision support value. Results: The eXtreme gradient boosting (XGB) model achieved the best performance, with an area under the receiver operating characteristic curve (AUROC) of 0.804 (95% confidence interval (CI) 0.792-0.816), area under the precision-recall curve (AUPRC) of 0.529 (95% CI 0.502-0.558) and a Brier score of 0.135 (95% CI 0.130-0.140). Discussion: The CIP cost curves provided a population-level overview of cost composition across decision thresholds, whereas LLM-generated cost-benefit analysis at individual patient-levels. The system was well received according to the evaluation by clinicians. However, feedback emphasizes the need to strengthen the technical accuracy for speculative tasks. Conclusion: CAP utilizes LLM agents to integrate ML classifier outcomes and cost-benefit analysis for more transparent and interpretable decision support.

Zhixing Li, Arsham Gholamzadeh Khoee, Yinan Yu

The objective of domain generalization (DG) is to enable models to be robust against domain shift. DG is crucial for deploying vision-language models (VLMs) in real-world applications, yet most existing methods rely on domain labels that may not be available and often ambiguous. We instead study the DG setting where models must generalize well without access to explicit domain labels. Our key idea is to represent an unseen target domain as a combination of latent domains automatically discovered from training data, enabling the model to adaptively transfer knowledge across domains. To realize this, we perform latent domain clustering on image features and fuse domain-specific text features based on the similarity between the input image and each latent domain. Experiments on four benchmarks show that this strategy yields consistent gains over VLM-based baselines and provides new insights into improving robustness under domain shift.

Yinan Yu, Alex Gonzalez-Caceres, Samuel Scheidegger et al.

Renovating existing buildings is essential for climate impact. Early-phase renovation planning requires simulations based on thermal 3D models at Level of Detail (LoD) 3, which include features like windows. However, scalable and accurate identification of such features remains a challenge. This paper presents the Scalable Image-to-3D Facade Parser (SI3FP), a pipeline that generates LoD3 thermal models by extracting geometries from images using both computer vision and deep learning. Unlike existing methods relying on segmentation and projection, SI3FP directly models geometric primitives in the orthographic image plane, providing a unified interface while reducing perspective distortions. SI3FP supports both sparse (e.g., Google Street View) and dense (e.g., hand-held camera) data sources. Tested on typical Swedish residential buildings, SI3FP achieved approximately 5% error in window-to-wall ratio estimates, demonstrating sufficient accuracy for early-stage renovation analysis. The pipeline facilitates large-scale energy renovation planning and has broader applications in urban development and planning.

Malsha Ashani Mahawatta Dona, Beatriz Cabrero-Daniel, Yinan Yu et al.

Large language models (LLMs) are growingly extended to process multimodal data such as text and video simultaneously. Their remarkable performance in understanding what is shown in images is surpassing specialized neural networks (NNs) such as Yolo that is supporting only a well-formed but very limited vocabulary, ie., objects that they are able to detect. When being non-restricted, LLMs and in particular state-of-the-art vision language models (VLMs) show impressive performance to describe even complex traffic situations. This is making them potentially suitable components for automotive perception systems to support the understanding of complex traffic situations or edge case situation. However, LLMs and VLMs are prone to hallucination, which mean to either potentially not seeing traffic agents such as vulnerable road users who are present in a situation, or to seeing traffic agents who are not there in reality. While the latter is unwanted making an ADAS or autonomous driving systems (ADS) to unnecessarily slow down, the former could lead to disastrous decisions from an ADS. In our work, we are systematically assessing the performance of 3 state-of-the-art VLMs on a diverse subset of traffic situations sampled from the Waymo Open Dataset to support safety guardrails for capturing such hallucinations in VLM-supported perception systems. We observe that both, proprietary and open VLMs exhibit remarkable image understanding capabilities even paying thorough attention to fine details sometimes difficult to spot for us humans. However, they are also still prone to making up elements in their descriptions to date requiring hallucination detection strategies such as BetterCheck that we propose in our work.

Yinan Yu, Samuel Scheidegger, Tomas McKelvey

Deep Convolutional Neural Networks (CNNs) have been widely used in various domains due to their impressive capabilities. These models are typically composed of a large number of 2D convolutional (Conv2D) layers with numerous trainable parameters. To reduce the complexity of a network, compression techniques can be applied. These methods typically rely on the analysis of trained deep learning models. However, in some applications, due to reasons such as particular data or system specifications and licensing restrictions, a pre-trained network may not be available. This would require the user to train a CNN from scratch. In this paper, we aim to find an alternative parameterization to Conv2D filters without relying on a pre-trained convolutional network. During the analysis, we observe that the effective rank of the vectorized Conv2D filters decreases with respect to the increasing depth in the network, which then leads to the implementation of the Depthwise Convolutional Eigen-Filter (DeCEF) layer. Essentially, a DeCEF layer is a low rank version of the Conv2D layer with significantly fewer trainable parameters and floating point operations (FLOPs). The way we define the effective rank is different from the previous work and it is easy to implement in any deep learning frameworks. To evaluate the effectiveness of DeCEF, experiments are conducted on the benchmark datasets CIFAR-10 and ImageNet using various network architectures. The results have shown a similar or higher accuracy and robustness using about 2/3 of the original parameters and reducing the number of FLOPs to 2/3 of the base network, which is then compared to the state-of-the-art techniques.

Yinan Yu, Tomas McKelvey

Metric learning for classification has been intensively studied over the last decade. The idea is to learn a metric space induced from a normed vector space on which data from different classes are well separated. Different measures of the separation thus lead to various designs of the objective function in the metric learning model. One classical metric is the Mahalanobis distance, where a linear transformation matrix is designed and applied on the original dataset to obtain a new subspace equipped with the Euclidean norm. The kernelized version has also been developed, followed by Multiple-Kernel learning models. In this paper, we consider metric learning to be the identification of the best kernel function with respect to a high class separability in the corresponding metric space. The contribution is twofold: 1) No pairwise computations are required as in most metric learning techniques; 2) Better flexibility and lower computational complexity is achieved using the CLAss-Specific (Multiple) Kernel - Metric Learning (CLAS(M)K-ML). The proposed techniques can be considered as a preprocessing step to any kernel method or kernel approximation technique. An extension to a hierarchical learning structure is also proposed to further improve the classification performance, where on each layer, the CLASMK is computed based on a selected "marginal" subset and feature vectors are constructed by concatenating the features from all previous layers.

Naiyu Gao, Yanhu Shan, Yupei Wang et al.

Recently, proposal-free instance segmentation has received increasing attention due to its concise and efficient pipeline. Generally, proposal-free methods generate instance-agnostic semantic segmentation labels and instance-aware features to group pixels into different object instances. However, previous methods mostly employ separate modules for these two sub-tasks and require multiple passes for inference. We argue that treating these two sub-tasks separately is suboptimal. In fact, employing multiple separate modules significantly reduces the potential for application. The mutual benefits between the two complementary sub-tasks are also unexplored. To this end, this work proposes a single-shot proposal-free instance segmentation method that requires only one single pass for prediction. Our method is based on a pixel-pair affinity pyramid, which computes the probability that two pixels belong to the same instance in a hierarchical manner. The affinity pyramid can also be jointly learned with the semantic class labeling and achieve mutual benefits. Moreover, incorporating with the learned affinity pyramid, a novel cascaded graph partition module is presented to sequentially generate instances from coarse to fine. Unlike previous time-consuming graph partition methods, this module achieves $5\times$ speedup and 9% relative improvement on Average-Precision (AP). Our approach achieves state-of-the-art results on the challenging Cityscapes dataset.

Jingchu Liu, Pengfei Hou, Lisen Mu et al.

Tactical driving decision making is crucial for autonomous driving systems and has attracted considerable interest in recent years. In this paper, we propose several practical components that can speed up deep reinforcement learning algorithms towards tactical decision making tasks: 1) non-uniform action skipping as a more stable alternative to action-repetition frame skipping, 2) a counter-based penalty for lanes on which ego vehicle has less right-of-road, and 3) heuristic inference-time action masking for apparently undesirable actions. We evaluate the proposed components in a realistic driving simulator and compare them with several baselines. Results show that the proposed scheme provides superior performance in terms of safety, efficiency, and comfort.

Yiyi Liao, Lichao Huang, Yue Wang et al.

Many standard robotic platforms are equipped with at least a fixed 2D laser range finder and a monocular camera. Although those platforms do not have sensors for 3D depth sensing capability, knowledge of depth is an essential part in many robotics activities. Therefore, recently, there is an increasing interest in depth estimation using monocular images. As this task is inherently ambiguous, the data-driven estimated depth might be unreliable in robotics applications. In this paper, we have attempted to improve the precision of monocular depth estimation by introducing 2D planar observation from the remaining laser range finder without extra cost. Specifically, we construct a dense reference map from the sparse laser range data, redefining the depth estimation task as estimating the distance between the real and the reference depth. To solve the problem, we construct a novel residual of residual neural network, and tightly combine the classification and regression losses for continuous depth estimation. Experimental results suggest that our method achieves considerable promotion compared to the state-of-the-art methods on both NYUD2 and KITTI, validating the effectiveness of our method on leveraging the additional sensory information. We further demonstrate the potential usage of our method in obstacle avoidance where our methodology provides comprehensive depth information compared to the solution using monocular camera or 2D laser range finder alone.

Lichao Huang, Yi Yang, Yafeng Deng et al.

How can a single fully convolutional neural network (FCN) perform on object detection? We introduce DenseBox, a unified end-to-end FCN framework that directly predicts bounding boxes and object class confidences through all locations and scales of an image. Our contribution is two-fold. First, we show that a single FCN, if designed and optimized carefully, can detect multiple different objects extremely accurately and efficiently. Second, we show that when incorporating with landmark localization during multi-task learning, DenseBox further improves object detection accuray. We present experimental results on public benchmark datasets including MALF face detection and KITTI car detection, that indicate our DenseBox is the state-of-the-art system for detecting challenging objects such as faces and cars.