Abstract
In this paper, we assess the movement error of a targeting system given target location data from artificial intelligence (AI) methods in automatic target recognition (ATR) systems. Few studies evaluate the impacts on the accuracy in moving a targeting system to an aimpoint provided in this manner. To address this knowledge gap, we assess the performance of a controlled gun turret system given target location from an object detector developed from AI methods. In our assessment, we define a measure of object detector error and examine the correlations with several standard metrics in object detection. We then statistically analyze the object detector error data and turret movement error data acquired from controlled targeting simulations, as well as their aggregate error, to examine the impact on turret movement accuracy. Finally, we study the correlations between additional metrics and the probability of a hit. The results indicate that AI technologies are a significant source of error to targeting systems. Moreover, the results suggest that metrics such as the confidence score, intersection-over-union, average precision and average recall are predictors of accuracy against stationary targets with our system parameters.
Title:
MakeWay: Object-Aware Costmaps for Proactive Indoor Navigation Using LiDAR
Authors: Binbin Xu, Allen Tao, Hugues Thomas, Jian Zhang, Timothy D. Barfoot
Abstract
In this paper, we introduce a LiDAR-based robot navigation system, based on novel object-aware affordance-based costmaps. Utilizing a 3D object detection network, our system identifies objects of interest in LiDAR keyframes, refines their 3D poses with the Iterative Closest Point (ICP) algorithm, and tracks them via Kalman filters and the Hungarian algorithm for data association. It then updates existing object poses with new associated detections and creates new object maps for unmatched detections. Using the maintained object-level mapping system, our system creates affordance-driven object costmaps for proactive collision avoidance in path planning. Additionally, we address the scarcity of indoor semantic LiDAR data by introducing an automated labeling technique. This method utilizes a CAD model database for accurate ground-truth annotations, encompassing bounding boxes, positions, orientations, and point-wise semantics of each object in LiDAR sequences. Our extensive evaluations, conducted in both simulated and real-world robot platforms, highlights the effectiveness of proactive object avoidance by using object affordance costmaps, enhancing robotic navigation safety and efficiency. The system can operate in real-time onboard and we intend to release our code and data for public use.
Title:
CP-VoteNet: Contrastive Prototypical VoteNet for Few-Shot Point Cloud Object Detection
Authors: Xuejing Li, Weijia Zhang, Chao Ma
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
Few-shot point cloud 3D object detection (FS3D) aims to identify and localise objects of novel classes from point clouds, using knowledge learnt from annotated base classes and novel classes with very few annotations. Thus far, this challenging task has been approached using prototype learning, but the performance remains far from satisfactory. We find that in existing methods, the prototypes are only loosely constrained and lack of fine-grained awareness of the semantic and geometrical correlation embedded within the point cloud space. To mitigate these issues, we propose to leverage the inherent contrastive relationship within the semantic and geometrical subspaces to learn more refined and generalisable prototypical representations. To this end, we first introduce contrastive semantics mining, which enables the network to extract discriminative categorical features by constructing positive and negative pairs within training batches. Meanwhile, since point features representing local patterns can be clustered into geometric components, we further propose to impose contrastive relationship at the primitive level. Through refined primitive geometric structures, the transferability of feature encoding from base to novel classes is significantly enhanced. The above designs and insights lead to our novel Contrastive Prototypical VoteNet (CP-VoteNet). Extensive experiments on two FS3D benchmarks FS-ScanNet and FS-SUNRGBD demonstrate that CP-VoteNet surpasses current state-of-the-art methods by considerable margins across different FS3D settings. Further ablation studies conducted corroborate the rationale and effectiveness of our designs.
Title:
PIB: Prioritized Information Bottleneck Framework for Collaborative Edge Video Analytics
Abstract
Collaborative edge sensing systems, particularly in collaborative perception systems in autonomous driving, can significantly enhance tracking accuracy and reduce blind spots with multi-view sensing capabilities. However, their limited channel capacity and the redundancy in sensory data pose significant challenges, affecting the performance of collaborative inference tasks. To tackle these issues, we introduce a Prioritized Information Bottleneck (PIB) framework for collaborative edge video analytics. We first propose a priority-based inference mechanism that jointly considers the signal-to-noise ratio (SNR) and the camera's coverage area of the region of interest (RoI). To enable efficient inference, PIB reduces video redundancy in both spatial and temporal domains and transmits only the essential information for the downstream inference tasks. This eliminates the need to reconstruct videos on the edge server while maintaining low latency. Specifically, it derives compact, task-relevant features by employing the deterministic information bottleneck (IB) method, which strikes a balance between feature informativeness and communication costs. Given the computational challenges caused by IB-based objectives with high-dimensional data, we resort to variational approximations for feasible optimization. Compared to TOCOM-TEM, JPEG, and HEVC, PIB achieves an improvement of up to 15.1\% in mean object detection accuracy (MODA) and reduces communication costs by 66.7% when edge cameras experience poor channel conditions.
Title:
UTrack: Multi-Object Tracking with Uncertain Detections
Authors: Edgardo Solano-Carrillo, Felix Sattler, Antje Alex, Alexander Klein, Bruno Pereira Costa, Angel Bueno Rodriguez, Jannis Stoppe
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
The tracking-by-detection paradigm is the mainstream in multi-object tracking, associating tracks to the predictions of an object detector. Although exhibiting uncertainty through a confidence score, these predictions do not capture the entire variability of the inference process. For safety and security critical applications like autonomous driving, surveillance, etc., knowing this predictive uncertainty is essential though. Therefore, we introduce, for the first time, a fast way to obtain the empirical predictive distribution during object detection and incorporate that knowledge in multi-object tracking. Our mechanism can easily be integrated into state-of-the-art trackers, enabling them to fully exploit the uncertainty in the detections. Additionally, novel association methods are introduced that leverage the proposed mechanism. We demonstrate the effectiveness of our contribution on a variety of benchmarks, such as MOT17, MOT20, DanceTrack, and KITTI.
Title:
Hybrid Classification-Regression Adaptive Loss for Dense Object Detection
Authors: Yanquan Huang, Liu Wei Zhen, Yun Hao, Mengyuan Zhang, Qingyao Wu, Zikun Deng, Xueming Liu, Hong Deng
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
For object detection detectors, enhancing model performance hinges on the ability to simultaneously consider inconsistencies across tasks and focus on difficult-to-train samples. Achieving this necessitates incorporating information from both the classification and regression tasks. However, prior work tends to either emphasize difficult-to-train samples within their respective tasks or simply compute classification scores with IoU, often leading to suboptimal model performance. In this paper, we propose a Hybrid Classification-Regression Adaptive Loss, termed as HCRAL. Specifically, we introduce the Residual of Classification and IoU (RCI) module for cross-task supervision, addressing task inconsistencies, and the Conditioning Factor (CF) to focus on difficult-to-train samples within each task. Furthermore, we introduce a new strategy named Expanded Adaptive Training Sample Selection (EATSS) to provide additional samples that exhibit classification and regression inconsistencies. To validate the effectiveness of the proposed method, we conduct extensive experiments on COCO test-dev. Experimental evaluations demonstrate the superiority of our approachs. Additionally, we designed experiments by separately combining the classification and regression loss with regular loss functions in popular one-stage models, demonstrating improved performance.
Title:
Structuring a Training Strategy to Robustify Perception Models with Realistic Image Augmentations
Authors: Ahmed Hammam, Bharathwaj Krishnaswami Sreedhar, Nura Kawa, Tim Patzelt, Oliver De Candido
Abstract
Advancing Machine Learning (ML)-based perception models for autonomous systems necessitates addressing weak spots within the models, particularly in challenging Operational Design Domains (ODDs). These are environmental operating conditions of an autonomous vehicle which can contain difficult conditions, e.g., lens flare at night or objects reflected in a wet street. This report introduces a novel methodology for training with augmentations to enhance model robustness and performance in such conditions. The proposed approach leverages customized physics-based augmentation functions, to generate realistic training data that simulates diverse ODD scenarios. We present a comprehensive framework that includes identifying weak spots in ML models, selecting suitable augmentations, and devising effective training strategies. The methodology integrates hyperparameter optimization and latent space optimization to fine-tune augmentation parameters, ensuring they maximally improve the ML models' performance. Experimental results demonstrate improvements in model performance, as measured by commonly used metrics such as mean Average Precision (mAP) and mean Intersection over Union (mIoU) on open-source object detection and semantic segmentation models and datasets. Our findings emphasize that optimal training strategies are model- and data-specific and highlight the benefits of integrating augmentations into the training pipeline. By incorporating augmentations, we observe enhanced robustness of ML-based perception models, making them more resilient to edge cases encountered in real-world ODDs. This work underlines the importance of customized augmentations and offers an effective solution for improving the safety and reliability of autonomous driving functions.
Keyword: transformer
Title:
HLogformer: A Hierarchical Transformer for Representing Log Data
Authors: Zhichao Hou, Mina Ghashami, Mikhail Kuznetsov, MohamadAli Torkamani
Abstract
Transformers have gained widespread acclaim for their versatility in handling diverse data structures, yet their application to log data remains underexplored. Log data, characterized by its hierarchical, dictionary-like structure, poses unique challenges when processed using conventional transformer models. Traditional methods often rely on manually crafted templates for parsing logs, a process that is labor-intensive and lacks generalizability. Additionally, the linear treatment of log sequences by standard transformers neglects the rich, nested relationships within log entries, leading to suboptimal representations and excessive memory usage. To address these issues, we introduce HLogformer, a novel hierarchical transformer framework specifically designed for log data. HLogformer leverages the hierarchical structure of log entries to significantly reduce memory costs and enhance representation learning. Unlike traditional models that treat log data as flat sequences, our framework processes log entries in a manner that respects their inherent hierarchical organization. This approach ensures comprehensive encoding of both fine-grained details and broader contextual relationships. Our contributions are threefold: First, HLogformer is the first framework to design a dynamic hierarchical transformer tailored for dictionary-like log data. Second, it dramatically reduces memory costs associated with processing extensive log sequences. Third, comprehensive experiments demonstrate that HLogformer more effectively encodes hierarchical contextual information, proving to be highly effective for downstream tasks such as synthetic anomaly detection and product recommendation.
Title:
Tex-ViT: A Generalizable, Robust, Texture-based dual-branch cross-attention deepfake detector
Abstract
Deepfakes, which employ GAN to produce highly realistic facial modification, are widely regarded as the prevailing method. Traditional CNN have been able to identify bogus media, but they struggle to perform well on different datasets and are vulnerable to adversarial attacks due to their lack of robustness. Vision transformers have demonstrated potential in the realm of image classification problems, but they require enough training data. Motivated by these limitations, this publication introduces Tex-ViT (Texture-Vision Transformer), which enhances CNN features by combining ResNet with a vision transformer. The model combines traditional ResNet features with a texture module that operates in parallel on sections of ResNet before each down-sampling operation. The texture module then serves as an input to the dual branch of the cross-attention vision transformer. It specifically focuses on improving the global texture module, which extracts feature map correlation. Empirical analysis reveals that fake images exhibit smooth textures that do not remain consistent over long distances in manipulations. Experiments were performed on different categories of FF++, such as DF, f2f, FS, and NT, together with other types of GAN datasets in cross-domain scenarios. Furthermore, experiments also conducted on FF++, DFDCPreview, and Celeb-DF dataset underwent several post-processing situations, such as blurring, compression, and noise. The model surpassed the most advanced models in terms of generalization, achieving a 98% accuracy in cross-domain scenarios. This demonstrates its ability to learn the shared distinguishing textural characteristics in the manipulated samples. These experiments provide evidence that the proposed model is capable of being applied to various situations and is resistant to many post-processing procedures.
Abstract
We present a limited empirical study of scaling laws for transfer learning in transformer models. More specifically, we examine a scaling law that incorporates a "transfer gap" term, indicating the effectiveness of pre-training on one distribution when optimizing for downstream performance on another distribution. When the transfer gap is low, pre-training is a cost-effective strategy for improving downstream performance. Conversely, when the gap is high, collecting high-quality fine-tuning data becomes relatively more cost effective. Fitting the scaling law to experiments from diverse datasets reveals significant variations in the transfer gap across distributions. In theory, the scaling law can inform optimal data allocation strategies and highlights how the scarcity of downstream data can bottleneck performance. Our findings contribute to a principled way to measure transfer learning efficiency and understand how data availability affects capabilities.
Title:
HiTSR: A Hierarchical Transformer for Reference-based Super-Resolution
Authors: Masoomeh Aslahishahri, Jordan Ubbens, Ian Stavness
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
In this paper, we propose HiTSR, a hierarchical transformer model for reference-based image super-resolution, which enhances low-resolution input images by learning matching correspondences from high-resolution reference images. Diverging from existing multi-network, multi-stage approaches, we streamline the architecture and training pipeline by incorporating the double attention block from GAN literature. Processing two visual streams independently, we fuse self-attention and cross-attention blocks through a gating attention strategy. The model integrates a squeeze-and-excitation module to capture global context from the input images, facilitating long-range spatial interactions within window-based attention blocks. Long skip connections between shallow and deep layers further enhance information flow. Our model demonstrates superior performance across three datasets including SUN80, Urban100, and Manga109. Specifically, on the SUN80 dataset, our model achieves PSNR/SSIM values of 30.24/0.821. These results underscore the effectiveness of attention mechanisms in reference-based image super-resolution. The transformer-based model attains state-of-the-art results without the need for purpose-built subnetworks, knowledge distillation, or multi-stage training, emphasizing the potency of attention in meeting reference-based image super-resolution requirements.
Title:
Training Ultra Long Context Language Model with Fully Pipelined Distributed Transformer
Authors: Jinghan Yao, Sam Ade Jacobs, Masahiro Tanaka, Olatunji Ruwase, Aamir Shafi, Hari Subramoni, Dhabaleswar K. Panda
Abstract
Large Language Models (LLMs) with long context capabilities are integral to complex tasks in natural language processing and computational biology, such as text generation and protein sequence analysis. However, training LLMs directly on extremely long contexts demands considerable GPU resources and increased memory, leading to higher costs and greater complexity. Alternative approaches that introduce long context capabilities via downstream finetuning or adaptations impose significant design limitations. In this paper, we propose Fully Pipelined Distributed Transformer (FPDT) for efficiently training long-context LLMs with extreme hardware efficiency. For GPT and Llama models, we achieve a 16x increase in sequence length that can be trained on the same hardware compared to current state-of-the-art solutions. With our dedicated sequence chunk pipeline design, we can now train 8B LLM with 2 million sequence length on only 4 GPUs, while also maintaining over 55% of MFU. Our proposed FPDT is agnostic to existing training techniques and is proven to work efficiently across different LLM models.
Title:
Cross Fusion RGB-T Tracking with Bi-directional Adapter
Abstract
Many state-of-the-art RGB-T trackers have achieved remarkable results through modality fusion. However, these trackers often either overlook temporal information or fail to fully utilize it, resulting in an ineffective balance between multi-modal and temporal information. To address this issue, we propose a novel Cross Fusion RGB-T Tracking architecture (CFBT) that ensures the full participation of multiple modalities in tracking while dynamically fusing temporal information. The effectiveness of CFBT relies on three newly designed cross spatio-temporal information fusion modules: Cross Spatio-Temporal Augmentation Fusion (CSTAF), Cross Spatio-Temporal Complementarity Fusion (CSTCF), and Dual-Stream Spatio-Temporal Adapter (DSTA). CSTAF employs a cross-attention mechanism to enhance the feature representation of the template comprehensively. CSTCF utilizes complementary information between different branches to enhance target features and suppress background features. DSTA adopts the adapter concept to adaptively fuse complementary information from multiple branches within the transformer layer, using the RGB modality as a medium. These ingenious fusions of multiple perspectives introduce only less than 0.3\% of the total modal parameters, but they indeed enable an efficient balance between multi-modal and temporal information. Extensive experiments on three popular RGB-T tracking benchmarks demonstrate that our method achieves new state-of-the-art performance.
Title:
Video to Music Moment Retrieval
Authors: Zijie Xin, Minquan Wang, Ye Ma, Bo Wang, Quan Chen, Peng Jiang, Xirong Li
Abstract
Adding proper background music helps complete a short video to be shared. Towards automating the task, previous research focuses on video-to-music retrieval (VMR), aiming to find amidst a collection of music the one best matching the content of a given video. Since music tracks are typically much longer than short videos, meaning the returned music has to be cut to a shorter moment, there is a clear gap between the practical need and VMR. In order to bridge the gap, we propose in this paper video to music moment retrieval (VMMR) as a new task. To tackle the new task, we build a comprehensive dataset Ad-Moment which contains 50K short videos annotated with music moments and develop a two-stage approach. In particular, given a test video, the most similar music is retrieved from a given collection. Then, a Transformer based music moment localization is performed. We term this approach Retrieval and Localization (ReaL). Extensive experiments on real-world datasets verify the effectiveness of the proposed method for VMMR.
Title:
BTMuda: A Bi-level Multi-source unsupervised domain adaptation framework for breast cancer diagnosis
Authors: Yuxiang Yang, Xinyi Zeng, Pinxian Zeng, Binyu Yan, Xi Wu, Jiliu Zhou, Yan Wang
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
Deep learning has revolutionized the early detection of breast cancer, resulting in a significant decrease in mortality rates. However, difficulties in obtaining annotations and huge variations in distribution between training sets and real scenes have limited their clinical applications. To address these limitations, unsupervised domain adaptation (UDA) methods have been used to transfer knowledge from one labeled source domain to the unlabeled target domain, yet these approaches suffer from severe domain shift issues and often ignore the potential benefits of leveraging multiple relevant sources in practical applications. To address these limitations, in this work, we construct a Three-Branch Mixed extractor and propose a Bi-level Multi-source unsupervised domain adaptation method called BTMuda for breast cancer diagnosis. Our method addresses the problems of domain shift by dividing domain shift issues into two levels: intra-domain and inter-domain. To reduce the intra-domain shift, we jointly train a CNN and a Transformer as two paths of a domain mixed feature extractor to obtain robust representations rich in both low-level local and high-level global information. As for the inter-domain shift, we redesign the Transformer delicately to a three-branch architecture with cross-attention and distillation, which learns domain-invariant representations from multiple domains. Besides, we introduce two alignment modules - one for feature alignment and one for classifier alignment - to improve the alignment process. Extensive experiments conducted on three public mammographic datasets demonstrate that our BTMuda outperforms state-of-the-art methods.
Title:
Vote&Mix: Plug-and-Play Token Reduction for Efficient Vision Transformer
Abstract
Despite the remarkable success of Vision Transformers (ViTs) in various visual tasks, they are often hindered by substantial computational cost. In this work, we introduce Vote\&Mix (\textbf{VoMix}), a plug-and-play and parameter-free token reduction method, which can be readily applied to off-the-shelf ViT models \textit{without any training}. VoMix tackles the computational redundancy of ViTs by identifying tokens with high homogeneity through a layer-wise token similarity voting mechanism. Subsequently, the selected tokens are mixed into the retained set, thereby preserving visual information. Experiments demonstrate VoMix significantly improves the speed-accuracy tradeoff of ViTs on both images and videos. Without any training, VoMix achieves a 2$\times$ increase in throughput of existing ViT-H on ImageNet-1K and a 2.4$\times$ increase in throughput of existing ViT-L on Kinetics-400 video dataset, with a mere 0.3\% drop in top-1 accuracy.
Title:
Stochastic Layer-Wise Shuffle: A Good Practice to Improve Vision Mamba Training
Authors: Zizheng Huang, Haoxing Chen, Jiaqi Li, Jun Lan, Huijia Zhu, Weiqiang Wang, Limin Wang
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
Recent Vision Mamba models not only have much lower complexity for processing higher resolution images and longer videos but also the competitive performance with Vision Transformers (ViTs). However, they are stuck into overfitting and thus only present up to base size (about 80M). It is still unclear how vanilla Vision Mamba (Vim) can be efficiently scaled up to larger sizes, which is essentially for further exploitation. In this paper, we propose a stochastic layer-wise shuffle regularization, which empowers successfully scaling non-hierarchical Vision Mamba to a large size (about 300M) in a supervised setting. Specifically, our base and large-scale ShuffleMamba models can outperform the supervised ViTs of similar size by 0.8\% and 1.0\% classification accuracy on ImageNet1k, respectively, without auxiliary data. When evaluated on the ADE20K semantic segmentation and COCO detection tasks, our ShuffleMamba models also show significant improvements. Without bells and whistles, the stochastic layer-wise shuffle has the following highlights: (1) \textit{Plug and play:} it does not change model architectures and will be omitted in inference. (2) \textit{Simple but effective:} it can improve the overfitting in Vim training and only introduce random token permutation operations. (3) \textit{Intuitive:} the token sequences in deeper layers are more likely to be shuffled as they are expected to be more semantic and less sensitive to patch positions. Code and models will be available at this https URL.
Title:
VQ4DiT: Efficient Post-Training Vector Quantization for Diffusion Transformers
Abstract
The Diffusion Transformers Models (DiTs) have transitioned the network architecture from traditional UNets to transformers, demonstrating exceptional capabilities in image generation. Although DiTs have been widely applied to high-definition video generation tasks, their large parameter size hinders inference on edge devices. Vector quantization (VQ) can decompose model weight into a codebook and assignments, allowing extreme weight quantization and significantly reducing memory usage. In this paper, we propose VQ4DiT, a fast post-training vector quantization method for DiTs. We found that traditional VQ methods calibrate only the codebook without calibrating the assignments. This leads to weight sub-vectors being incorrectly assigned to the same assignment, providing inconsistent gradients to the codebook and resulting in a suboptimal result. To address this challenge, VQ4DiT calculates the candidate assignment set for each weight sub-vector based on Euclidean distance and reconstructs the sub-vector based on the weighted average. Then, using the zero-data and block-wise calibration method, the optimal assignment from the set is efficiently selected while calibrating the codebook. VQ4DiT quantizes a DiT XL/2 model on a single NVIDIA A100 GPU within 20 minutes to 5 hours depending on the different quantization settings. Experiments show that VQ4DiT establishes a new state-of-the-art in model size and performance trade-offs, quantizing weights to 2-bit precision while retaining acceptable image generation quality.
Title:
Temporal and Interactive Modeling for Efficient Human-Human Motion Generation
Authors: Yabiao Wang, Shuo Wang, Jiangning Zhang, Ke Fan, Jiafu Wu, Zhengkai Jiang, Yong Liu
Subjects: Subjects:
Computer Vision and Pattern Recognition (cs.CV)
Abstract
Human-human motion generation is essential for understanding humans as social beings. Although several transformer-based methods have been proposed, they typically model each individual separately and overlook the causal relationships in temporal motion sequences. Furthermore, the attention mechanism in transformers exhibits quadratic computational complexity, significantly reducing their efficiency when processing long sequences. In this paper, we introduce TIM (Temporal and Interactive Modeling), an efficient and effective approach that presents the pioneering human-human motion generation model utilizing RWKV. Specifically, we first propose Causal Interactive Injection to leverage the temporal properties of motion sequences and avoid non-causal and cumbersome modeling. Then we present Role-Evolving Mixing to adjust to the ever-evolving roles throughout the interaction. Finally, to generate smoother and more rational motion, we design Localized Pattern Amplification to capture short-term motion patterns. Extensive experiments on InterHuman demonstrate that our method achieves superior performance. Notably, TIM has achieved state-of-the-art results using only 32% of InterGen's trainable parameters. Code will be available soon. Homepage: this https URL
Title:
The Iterative Optimal Brain Surgeon: Faster Sparse Recovery by Leveraging Second-Order Information
Abstract
The rising footprint of machine learning has led to a focus on imposing \emph{model sparsity} as a means of reducing computational and memory costs. For deep neural networks (DNNs), the state-of-the-art accuracy-vs-sparsity is achieved by heuristics inspired by the classical Optimal Brain Surgeon (OBS) framework~\citep{lecun90brain, hassibi1992second, hassibi1993optimal}, which leverages loss curvature information to make better pruning decisions. Yet, these results still lack a solid theoretical understanding, and it is unclear whether they can be improved by leveraging connections to the wealth of work on sparse recovery algorithms. In this paper, we draw new connections between these two areas and present new sparse recovery algorithms inspired by the OBS framework that comes with theoretical guarantees under reasonable assumptions and have strong practical performance. Specifically, our work starts from the observation that we can leverage curvature information in OBS-like fashion upon the projection step of classic iterative sparse recovery algorithms such as IHT. We show for the first time that this leads both to improved convergence bounds under standard assumptions. Furthermore, we present extensions of this approach to the practical task of obtaining accurate sparse DNNs, and validate it experimentally at scale for Transformer-based models on vision and language tasks.
Title:
Improving Extraction of Clinical Event Contextual Properties from Electronic Health Records: A Comparative Study
Authors: Shubham Agarwal, Thomas Searle, Mart Ratas, Anthony Shek, James Teo, Richard Dobson
Subjects: Subjects:
Computation and Language (cs.CL)
Abstract
Electronic Health Records are large repositories of valuable clinical data, with a significant portion stored in unstructured text format. This textual data includes clinical events (e.g., disorders, symptoms, findings, medications and procedures) in context that if extracted accurately at scale can unlock valuable downstream applications such as disease prediction. Using an existing Named Entity Recognition and Linking methodology, MedCAT, these identified concepts need to be further classified (contextualised) for their relevance to the patient, and their temporal and negated status for example, to be useful downstream. This study performs a comparative analysis of various natural language models for medical text classification. Extensive experimentation reveals the effectiveness of transformer-based language models, particularly BERT. When combined with class imbalance mitigation techniques, BERT outperforms Bi-LSTM models by up to 28% and the baseline BERT model by up to 16% for recall of the minority classes. The method has been implemented as part of CogStack/MedCAT framework and made available to the community for further research.
Title:
Investigating Neuron Ablation in Attention Heads: The Case for Peak Activation Centering
Authors: Nicholas Pochinkov, Ben Pasero, Skylar Shibayama
Subjects: Subjects:
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
Abstract
The use of transformer-based models is growing rapidly throughout society. With this growth, it is important to understand how they work, and in particular, how the attention mechanisms represent concepts. Though there are many interpretability methods, many look at models through their neuronal activations, which are poorly understood. We describe different lenses through which to view neuron activations, and investigate the effectiveness in language models and vision transformers through various methods of neural ablation: zero ablation, mean ablation, activation resampling, and a novel approach we term 'peak ablation'. Through experimental analysis, we find that in different regimes and models, each method can offer the lowest degradation of model performance compared to other methods, with resampling usually causing the most significant performance deterioration. We make our code available at this https URL.
Title:
Modularity in Transformers: Investigating Neuron Separability & Specialization
Authors: Nicholas Pochinkov, Thomas Jones, Mohammed Rashidur Rahman
Subjects: Subjects:
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
Abstract
Transformer models are increasingly prevalent in various applications, yet our understanding of their internal workings remains limited. This paper investigates the modularity and task specialization of neurons within transformer architectures, focusing on both vision (ViT) and language (Mistral 7B) models. Using a combination of selective pruning and MoEfication clustering techniques, we analyze the overlap and specialization of neurons across different tasks and data subsets. Our findings reveal evidence of task-specific neuron clusters, with varying degrees of overlap between related tasks. We observe that neuron importance patterns persist to some extent even in randomly initialized models, suggesting an inherent structure that training refines. Additionally, we find that neuron clusters identified through MoEfication correspond more strongly to task-specific neurons in earlier and later layers of the models. This work contributes to a more nuanced understanding of transformer internals and offers insights into potential avenues for improving model interpretability and efficiency.
Title:
Assessing Generative Language Models in Classification Tasks: Performance and Self-Evaluation Capabilities in the Environmental and Climate Change Domain
Authors: Francesca Grasso, Stefano Locci
Subjects: Subjects:
Computation and Language (cs.CL)
Abstract
This paper examines the performance of two Large Language Models (LLMs), GPT3.5 and Llama2 and one Small Language Model (SLM) Gemma, across three different classification tasks within the climate change (CC) and environmental domain. Employing BERT-based models as a baseline, we compare their efficacy against these transformer-based models. Additionally, we assess the models' self-evaluation capabilities by analyzing the calibration of verbalized confidence scores in these text classification tasks. Our findings reveal that while BERT-based models generally outperform both the LLMs and SLM, the performance of the large generative models is still noteworthy. Furthermore, our calibration analysis reveals that although Gemma is well-calibrated in initial tasks, it thereafter produces inconsistent results; Llama is reasonably calibrated, and GPT consistently exhibits strong calibration. Through this research, we aim to contribute to the ongoing discussion on the utility and effectiveness of generative LMs in addressing some of the planet's most urgent issues, highlighting their strengths and limitations in the context of ecology and CC.
Title:
CLOCR-C: Context Leveraging OCR Correction with Pre-trained Language Models
Authors: Jonathan Bourne
Subjects: Subjects:
Computation and Language (cs.CL); Digital Libraries (cs.DL)
Abstract
The digitisation of historical print media archives is crucial for increasing accessibility to contemporary records. However, the process of Optical Character Recognition (OCR) used to convert physical records to digital text is prone to errors, particularly in the case of newspapers and periodicals due to their complex layouts. This paper introduces Context Leveraging OCR Correction (CLOCR-C), which utilises the infilling and context-adaptive abilities of transformer-based language models (LMs) to improve OCR quality. The study aims to determine if LMs can perform post-OCR correction, improve downstream NLP tasks, and the value of providing the socio-cultural context as part of the correction process. Experiments were conducted using seven LMs on three datasets: the 19th Century Serials Edition (NCSE) and two datasets from the Overproof collection. The results demonstrate that some LMs can significantly reduce error rates, with the top-performing model achieving over a 60% reduction in character error rate on the NCSE dataset. The OCR improvements extend to downstream tasks, such as Named Entity Recognition, with increased Cosine Named Entity Similarity. Furthermore, the study shows that providing socio-cultural context in the prompts improves performance, while misleading prompts lower performance. In addition to the findings, this study releases a dataset of 91 transcribed articles from the NCSE, containing a total of 40 thousand words, to support further research in this area. The findings suggest that CLOCR-C is a promising approach for enhancing the quality of existing digital archives by leveraging the socio-cultural information embedded in the LMs and the text requiring correction.
Keyword: scene understanding
Title:
UrBench: A Comprehensive Benchmark for Evaluating Large Multimodal Models in Multi-View Urban Scenarios
Abstract
Recent evaluations of Large Multimodal Models (LMMs) have explored their capabilities in various domains, with only few benchmarks specifically focusing on urban environments. Moreover, existing urban benchmarks have been limited to evaluating LMMs with basic region-level urban tasks under singular views, leading to incomplete evaluations of LMMs' abilities in urban environments. To address these issues, we present UrBench, a comprehensive benchmark designed for evaluating LMMs in complex multi-view urban scenarios. UrBench contains 11.6K meticulously curated questions at both region-level and role-level that cover 4 task dimensions: Geo-Localization, Scene Reasoning, Scene Understanding, and Object Understanding, totaling 14 task types. In constructing UrBench, we utilize data from existing datasets and additionally collect data from 11 cities, creating new annotations using a cross-view detection-matching method. With these images and annotations, we then integrate LMM-based, rule-based, and human-based methods to construct large-scale high-quality questions. Our evaluations on 21 LMMs show that current LMMs struggle in the urban environments in several aspects. Even the best performing GPT-4o lags behind humans in most tasks, ranging from simple tasks such as counting to complex tasks such as orientation, localization and object attribute recognition, with an average performance gap of 17.4%. Our benchmark also reveals that LMMs exhibit inconsistent behaviors with different urban views, especially with respect to understanding cross-view relations. UrBench datasets and benchmark results will be publicly available at this https URL.
Keyword: human object interaction
There is no result
Keyword: visual relation detection
There is no result
Keyword: object detection
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Keyword: transformer
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Title:
Keyword: scene understanding
Title:
Keyword: visual reasoning
There is no result