Knowing When to Answer: Adaptive Confidence Refinement for Reliable Audio-Visual Question Answering

We present a formal problem formulation for \textit{Reliable} Audio-Visual Question Answering ($\mathcal{R}$-AVQA), where we prefer abstention over answering incorrectly. While recent AVQA models have high accuracy, their ability to identify when they are likely wrong and their consequent abstention from answering remain underexplored areas of research. To fill this gap, we explore several approaches and then propose Adaptive Confidence Refinement (ACR), a lightweight method to further enhance the performance of $\mathcal{R}$-AVQA. Our key insight is that the Maximum Softmax Probability (MSP) is Bayes-optimal only under strong calibration, a condition usually not met in deep neural networks, particularly in multimodal models. Instead of replacing MSP, our ACR maintains it as a primary confidence signal and applies input-adaptive residual corrections when MSP is deemed unreliable. ACR introduces two learned heads: i) a Residual Risk Head that predicts low-magnitude correctness residuals that MSP does not capture, and ii) a Confidence Gating Head to determine MSP trustworthiness. Our experiments and theoretical analysis show that ACR consistently outperforms existing methods on in- and out-of-disrtibution, and data bias settings across three different AVQA architectures, establishing a solid foundation for $\mathcal{R}$-AVQA task. The code and checkpoints will be available upon acceptance \href{https://github.com/PhuTran1005/R-AVQA}{at here}

Model-Dowser: Data-Free Importance Probing to Mitigate Catastrophic Forgetting in Multimodal Large Language Models

Fine-tuning Multimodal Large Language Models (MLLMs) on task-specific data is an effective way to improve performance on downstream applications. However, such adaptation often leads to a degradation in generalization on pretrained tasks, a phenomenon known as Catastrophic Forgetting. Existing methods that aim to mitigate this issue either become ineffective when fine-tuning deeper layers of the language decoder or scale poorly with increasing model size. To address these limitations, we propose Model-Dowser, a novel sparse fine-tuning approach for MLLMs. Model-Dowser measures a principled importance score for each model parameter with respect to pretrained generalization (prior to downstream adaptation) by jointly considering weight magnitudes, input activations, and output sensitivities. During fine-tuning, Model-Dowser selectively preserves high-importance parameters and updates the remaining. Comprehensive experiments on two representative MLLMs, LLaVA and NVILA, demonstrate that Model-Dowser effectively mitigates catastrophic forgetting and consistently outperforms prior methods, while remaining resource-efficient and scalable to multi-billion-parameter models.

LungCRCT: Causal Representation based Lung CT Processing for Lung Cancer Treatment

Due to silence in early stages, lung cancer has been one of the most leading causes of mortality in cancer patients world-wide. Moreover, major symptoms of lung cancer are hard to differentiate with other respiratory disease symptoms such as COPD, further leading patients to overlook cancer progression in early stages. Thus, to enhance survival rates in lung cancer, early detection from consistent proactive respiratory system monitoring becomes crucial. One of the most prevalent and effective methods for lung cancer monitoring would be low-dose computed tomography(LDCT) chest scans, which led to remarkable enhancements in lung cancer detection or tumor classification tasks under rapid advancements and applications of computer vision based AI models such as EfficientNet or ResNet in image processing. However, though advanced CNN models under transfer learning or ViT based models led to high performing lung cancer detections, due to its intrinsic limitations in terms of correlation dependence and low interpretability due to complexity, expansions of deep learning models to lung cancer treatment analysis or causal intervention analysis simulations are still limited. Therefore, this research introduced LungCRCT: a latent causal representation learning based lung cancer analysis framework that retrieves causal representations of factors within the physical causal mechanism of lung cancer progression. With the use of advanced graph autoencoder based causal discovery algorithms with distance Correlation disentanglement and entropy-based image reconstruction refinement, LungCRCT not only enables causal intervention analysis for lung cancer treatments, but also leads to robust, yet extremely light downstream models in malignant tumor classification tasks with an AUC score of 93.91%.

Vision Language Models are Biased

Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that help them on downstream tasks but also may notoriously sway their outputs towards wrong or biased answers. In this work, we test how the knowledge about popular subjects hurt the accuracy of vision language models (VLMs) on standard, objective visual tasks of counting and identification. We find that state-of-the-art VLMs are strongly biased (e.g, unable to recognize a fourth stripe has been added to a 3-stripe Adidas logo) scoring an average of 17.05% accuracy in counting (e.g., counting stripes in an Adidas-like logo) across 7 diverse domains from animals, logos, chess, board games, optical illusions, to patterned grids. Insert text (e.g., "Adidas") describing the subject name into the counterfactual image further decreases VLM accuracy. The biases in VLMs are so strong that instructing them to double-check their results or rely exclusively on image details to answer improves counting accuracy by only +2 points, on average. Our work presents an interesting failure mode in VLMs and an automated framework for testing VLM biases. Code and data are available at: vlmsarebiased.github.io.

B-score: Detecting biases in large language models using response history

Large language models (LLMs) often exhibit strong biases, e.g, against women or in favor of the number 7. We investigate whether LLMs would be able to output less biased answers when allowed to observe their prior answers to the same question in a multi-turn conversation. To understand which types of questions invite more biased answers, we test LLMs on our proposed set of questions that span 9 topics and belong to three types: (1) Subjective; (2) Random; and (3) Objective. Interestingly, LLMs are able to "de-bias" themselves in a multi-turn conversation in response to questions that seek an Random, unbiased answer. Furthermore, we propose B-score, a novel metric that is effective in detecting biases to Subjective, Random, Easy, and Hard questions. On MMLU, HLE, and CSQA, leveraging B-score substantially improves the verification accuracy of LLM answers (i.e, accepting LLM correct answers and rejecting incorrect ones) compared to using verbalized confidence scores or the frequency of single-turn answers alone. Code and data are available at: https://b-score.github.io.

Rethinking the Nested U-Net Approach: Enhancing Biomarker Segmentation with Attention Mechanisms and Multiscale Feature Fusion

Identifying biomarkers in medical images is vital for a wide range of biotech applications. However, recent Transformer and CNN based methods often struggle with variations in morphology and staining, which limits their feature extraction capabilities. In medical image segmentation, where data samples are often limited, state-of-the-art (SOTA) methods improve accuracy by using pre-trained encoders, while end-to-end approaches typically fall short due to difficulties in transferring multiscale features effectively between encoders and decoders. To handle these challenges, we introduce a nested UNet architecture that captures both local and global context through Multiscale Feature Fusion and Attention Mechanisms. This design improves feature integration from encoders, highlights key channels and regions, and restores spatial details to enhance segmentation performance. Our method surpasses SOTA approaches, as evidenced by experiments across four datasets and detailed ablation studies. Code: https://github.com/saadwazir/ReN-UNet

QuantuneV2: Compiler-Based Local Metric-Driven Mixed Precision Quantization for Practical Embedded AI Applications

Mixed-precision quantization methods have been proposed to reduce model size while minimizing accuracy degradation. However, existing studies require retraining and do not consider the computational overhead and intermediate representations (IR) generated during the compilation process, limiting their application at the compiler level. This computational overhead refers to the runtime latency caused by frequent quantization and dequantization operations during inference. Performing these operations at the individual operator level causes significant runtime delays. To address these issues, we propose QuantuneV2, a compiler-based mixed-precision quantization method designed for practical embedded AI applications. QuantuneV2 performs inference only twice, once before quantization and once after quantization, and operates with a computational complexity of O(n) that increases linearly with the number of model parameters. We also made the sensitivity analysis more stable by using local metrics like weights, activation values, the Signal to Quantization Noise Ratio, and the Mean Squared Error. We also cut down on computational overhead by choosing the best IR and using operator fusion. Experimental results show that QuantuneV2 achieved up to a 10.28 percent improvement in accuracy and a 12.52 percent increase in speed compared to existing methods across five models: ResNet18v1, ResNet50v1, SqueezeNetv1, VGGNet, and MobileNetv2. This demonstrates that QuantuneV2 enhances model performance while maintaining computational efficiency, making it suitable for deployment in embedded AI environments.

VARCO-VISION: Expanding Frontiers in Korean Vision-Language Models

In this paper, we introduce an open-source Korean-English vision-language model (VLM), VARCO-VISION. We incorporate a step-by-step training strategy that allows a model learn both linguistic and visual information while preserving the backbone model's knowledge. Our model demonstrates outstanding performance in diverse settings requiring bilingual image-text understanding and generation abilities compared to models of similar size. VARCO-VISION is also capable of grounding, referring, and OCR, expanding its usage and potential applications for real-world scenarios. In addition to the model, we release five Korean evaluation datasets, including four closed-set and one openset benchmarks. We anticipate that our milestone will broaden the opportunities for AI researchers aiming to train VLMs. VARCO-VISION is available at https://huggingface.co/NCSOFT/VARCO-VISION-14B.

Channel-Partitioned Windowed Attention And Frequency Learning for Single Image Super-Resolution

Recently, window-based attention methods have shown great potential for computer vision tasks, particularly in Single Image Super-Resolution (SISR). However, it may fall short in capturing long-range dependencies and relationships between distant tokens. Additionally, we find that learning on spatial domain does not convey the frequency content of the image, which is a crucial aspect in SISR. To tackle these issues, we propose a new Channel-Partitioned Attention Transformer (CPAT) to better capture long-range dependencies by sequentially expanding windows along the height and width of feature maps. In addition, we propose a novel Spatial-Frequency Interaction Module (SFIM), which incorporates information from spatial and frequency domains to provide a more comprehensive information from feature maps. This includes information about the frequency content and enhances the receptive field across the entire image. Experimental findings demonstrate the effectiveness of our proposed modules and architecture. In particular, CPAT surpasses current state-of-the-art methods by up to 0.31dB.

OffsetBias: Leveraging Debiased Data for Tuning Evaluators

Employing Large Language Models (LLMs) to assess the quality of generated responses, such as prompting instruct-tuned models or fine-tuning judge models, has become a widely adopted evaluation method. It is also known that such evaluators are vulnerable to biases, such as favoring longer responses. While it is important to overcome this problem, the specifics of these biases remain under-explored. In this work, we qualitatively identify six types of biases inherent in various judge models. We propose EvalBiasBench as a meta-evaluation collection of hand-crafted test cases for each bias type. Additionally, we present de-biasing dataset construction methods and the associated preference dataset OffsetBias. Experimental results demonstrate that fine-tuning on our dataset significantly enhances the robustness of judge models against biases and improves performance across most evaluation scenarios. We release our datasets and the fine-tuned judge model to public.