arXiv subjects

Wonju Lee

Publications and source records attributed to Wonju Lee.

7 recordsLinked to original sources

arXiv2026

Symmetry-Aware Fusion of Vision and Tactile Sensing via Bilateral Force Priors for Robotic Manipulation

Insertion tasks in robotic manipulation demand precise, contact-rich interactions that vision alone cannot resolve. While tactile feedback is intuitively valuable, existing studies have shown that naïve visuo-tactile fusion often fails to deliver consistent improvements. In this work, we propose a Cross-Modal Transformer (CMT) for visuo-tactile fusion that integrates wrist-camera observations with tactile signals through structured self- and cross-attention. To stabilize tactile embeddings, we further introduce a physics-informed regularization that encourages bilateral force balance, reflecting principles of human motor control. Experiments on the TacSL benchmark show that CMT with symmetry regularization achieves a 96.59% insertion success rate, surpassing naïve and gated fusion baselines and closely matching the privileged "wrist + contact force" configuration (96.09%). These results highlight two central insights: (i) tactile sensing is indispensable for precise alignment, and (ii) principled multimodal fusion, further strengthened by physics-informed regularization, unlocks complementary strengths of vision and touch, approaching privileged performance under realistic sensing.

cs.RO↗

arXiv2023

ViT-ReciproCAM: Gradient and Attention-Free Visual Explanations for Vision Transformer

This paper presents a novel approach to address the challenges of understanding the prediction process and debugging prediction errors in Vision Transformers (ViT), which have demonstrated superior performance in various computer vision tasks such as image classification and object detection. While several visual explainability techniques, such as CAM, Grad-CAM, Score-CAM, and Recipro-CAM, have been extensively researched for Convolutional Neural Networks (CNNs), limited research has been conducted on ViT. Current state-of-the-art solutions for ViT rely on class agnostic Attention-Rollout and Relevance techniques. In this work, we propose a new gradient-free visual explanation method for ViT, called ViT-ReciproCAM, which does not require attention matrix and gradient information. ViT-ReciproCAM utilizes token masking and generated new layer outputs from the target layer's input to exploit the correlation between activated tokens and network predictions for target classes. Our proposed method outperforms the state-of-the-art Relevance method in the Average Drop-Coherence-Complexity (ADCC) metric by $4.58\%$ to $5.80\%$ and generates more localized saliency maps. Our experiments demonstrate the effectiveness of ViT-ReciproCAM and showcase its potential for understanding and debugging ViT models. Our proposed method provides an efficient and easy-to-implement alternative for generating visual explanations, without requiring attention and gradient information, which can be beneficial for various applications in the field of computer vision.

cs.CV↗

arXiv2023

Recipro-CAM: Fast gradient-free visual explanations for convolutional neural networks

The Convolutional Neural Network (CNN) is a widely used deep learning architecture for computer vision. However, its black box nature makes it difficult to interpret the behavior of the model. To mitigate this issue, AI practitioners have explored explainable AI methods like Class Activation Map (CAM) and Grad-CAM. Although these methods have shown promise, they are limited by architectural constraints or the burden of gradient computing. To overcome this issue, Score-CAM and Ablation-CAM have been proposed as gradient-free methods, but they have longer execution times compared to CAM or Grad-CAM based methods, making them unsuitable for real-world solution though they resolved gradient related issues and enabled inference mode XAI. To address this challenge, we propose a fast gradient-free Reciprocal CAM (Recipro-CAM) method. Our approach involves spatially masking the extracted feature maps to exploit the correlation between activation maps and network predictions for target classes. Our proposed method has yielded promising results, outperforming current state-of-the-art method in the Average Drop-Coherence-Complexity (ADCC) metric by $1.78 \%$ to $3.72 \%$, excluding VGG-16 backbone. Moreover, Recipro-CAM generates saliency maps at a similar rate to Grad-CAM and is approximately $148$ times faster than Score-CAM. The source code for Recipro-CAM is available in our data analysis framework.

cs.CV↗

arXiv2021

Unsupervised Model Drift Estimation with Batch Normalization Statistics for Dataset Shift Detection and Model Selection

While many real-world data streams imply that they change frequently in a nonstationary way, most of deep learning methods optimize neural networks on training data, and this leads to severe performance degradation when dataset shift happens. However, it is less possible to annotate or inspect newly streamed data by humans, and thus it is desired to measure model drift at inference time in an unsupervised manner. In this paper, we propose a novel method of model drift estimation by exploiting statistics of batch normalization layer on unlabeled test data. To remedy possible sampling error of streamed input data, we adopt low-rank approximation to each representational layer. We show the effectiveness of our method not only on dataset shift detection but also on model selection when there are multiple candidate models among model zoo or training trajectories in an unsupervised way. We further demonstrate the consistency of our method by comparing model drift scores between different network architectures.

cs.LG↗

arXiv2017

Coded Multicast Fronthauling and Edge Caching for Multi-Connectivity Transmission in Fog Radio Access Networks

This work studies the advantages of coded multicasting for the downlink of a Fog Radio Access Network (F-RAN) system equipped with a multicast fronthaul link. In this system, a control unit (CU) in the baseband processing unit (BBU) pool is connected to distributed edge nodes (ENs) through a multicast fronthaul link of finite capacity, and the ENs have baseband processing and caching capabilities. Each user equipment (UE) requests a file in a content library which is available at the CU, and the requested files are served by the closest ENs based on the cached contents and on the information received on the multicast fronthaul link. The performance of coded multicast fronthauling is investigated in terms of the delivery latency of the requested contents under the assumption of pipelined transmission on the fronthaul and edge links and of single-user encoding and decoding strategies based on the hard transfer of files on the fronthaul links. Extensive numerical results are provided to validate the advantages of the coded multicasting scheme compared to uncoded unicast and multicast strategies.

cs.IT↗

arXiv2016

Multivariate Fronthaul Quantization for Downlink C-RAN

The Cloud-Radio Access Network (C-RAN) cellular architecture relies on the transfer of complex baseband signals to and from a central unit (CU) over digital fronthaul links to enable the virtualization of the baseband processing functionalities of distributed radio units (RUs). The standard design of digital fronthauling is based on either scalar quantization or on more sophisticated point to-point compression techniques operating on baseband signals. Motivated by network-information theoretic results, techniques for fronthaul quantization and compression that improve over point-to-point solutions by allowing for joint processing across multiple fronthaul links at the CU have been recently proposed for both the uplink and the downlink. For the downlink, a form of joint compression, known in network information theory as multivariate compression, was shown to be advantageous under a non-constructive asymptotic information-theoretic framework. In this paper, instead, the design of a practical symbol-by-symbol fronthaul quantization algorithm that implements the idea of multivariate compression is investigated for the C-RAN downlink. As compared to current standards, the proposed multivariate quantization (MQ) only requires changes in the CU processing while no modification is needed at the RUs. The algorithm is extended to enable the joint optimization of downlink precoding and quantization, reduced-complexity MQ via successive block quantization, and variable-length compression. Numerical results, which include performance evaluations over standard cellular models, demonstrate the advantages of MQ and the merits of a joint optimization with precoding.

cs.IT↗

arXiv2015

HARQ Buffer Management: An Information-Theoretic View

A key practical constraint on the design of Hybrid automatic repeat request (HARQ) schemes is the size of the on-chip buffer that is available at the receiver to store previously received packets. In fact, in modern wireless standards such as LTE and LTE-A, the HARQ buffer size is one of the main drivers of the modem area and power consumption. This has recently highlighted the importance of HARQ buffer management, that is, of the use of buffer-aware transmission schemes and of advanced compression policies for the storage of received data. This work investigates HARQ buffer management by leveraging information-theoretic achievability arguments based on random coding. Specifically, standard HARQ schemes, namely Type-I, Chase Combining and Incremental Redundancy, are first studied under the assumption of a finite-capacity HARQ buffer by considering both coded modulation, via Gaussian signaling, and Bit Interleaved Coded Modulation (BICM). The analysis sheds light on the impact of different compression strategies, namely the conventional compression log-likelihood ratios and the direct digitization of baseband signals, on the throughput. Then, coding strategies based on layered modulation and optimized coding blocklength are investigated, highlighting the benefits of HARQ buffer-aware transmission schemes. The optimization of baseband compression for multiple-antenna links is also studied, demonstrating the optimality of a transform coding approach.

cs.IT↗