Abstract
With the ever-growing model size and the limited availability of labeled training data, transfer learning has become an increasingly popular approach in many science and engineering domains. For classification problems, this work delves into the mystery of transfer learning through an intriguing phenomenon termed neural collapse (NC), where the last-layer features and classifiers of learned deep networks satisfy: (i) the within-class variability of the features collapses to zero, and (ii) the between-class feature means are maximally and equally separated. Through the lens of NC, our findings for transfer learning are the following: (i) when pre-training models, preventing intra-class variability collapse (to a certain extent) better preserves the intrinsic structures of the input data, so that it leads to better model transferability; (ii) when fine-tuning models on downstream tasks, obtaining features with more NC on downstream data results in better test accuracy on the given task. The above results not only demystify many widely used heuristics in model pre-training (e.g., data augmentation, projection head, self-supervised learning), but also leads to more efficient and principled fine-tuning method on downstream tasks that we demonstrate through extensive experimental results.
Benchmark for Uncertainty & Robustness in Self-Supervised Learning
Authors: Ha Manh Bui, Iliana Maifeld-Carucci
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
Self-Supervised Learning (SSL) is crucial for real-world applications, especially in data-hungry domains such as healthcare and self-driving cars. In addition to a lack of labeled data, these applications also suffer from distributional shifts. Therefore, an SSL method should provide robust generalization and uncertainty estimation in the test dataset to be considered a reliable model in such high-stakes domains. However, existing approaches often focus on generalization, without evaluating the model's uncertainty. The ability to compare SSL techniques for improving these estimates is therefore critical for research on the reliability of self-supervision models. In this paper, we explore variants of SSL methods, including Jigsaw Puzzles, Context, Rotation, Geometric Transformations Prediction for vision, as well as BERT and GPT for language tasks. We train SSL in auxiliary learning for vision and pre-training for language model, then evaluate the generalization (in-out classification accuracy) and uncertainty (expected calibration error) across different distribution covariate shift datasets, including MNIST-C, CIFAR-10-C, CIFAR-10.1, and MNLI. Our goal is to create a benchmark with outputs from experiments, providing a starting point for new SSL methods in Reliable Machine Learning. All source code to reproduce results is available at https://github.com/hamanhbui/reliable_ssl_baselines.
Adaptive Risk-Aware Bidding with Budget Constraint in Display Advertising
Abstract
Real-time bidding (RTB) has become a major paradigm of display advertising. Each ad impression generated from a user visit is auctioned in real time, where demand-side platform (DSP) automatically provides bid price usually relying on the ad impression value estimation and the optimal bid price determination. However, the current bid strategy overlooks large randomness of the user behaviors (e.g., click) and the cost uncertainty caused by the auction competition. In this work, we explicitly factor in the uncertainty of estimated ad impression values and model the risk preference of a DSP under a specific state and market environment via a sequential decision process. Specifically, we propose a novel adaptive risk-aware bidding algorithm with budget constraint via reinforcement learning, which is the first to simultaneously consider estimation uncertainty and the dynamic risk tendency of a DSP. We theoretically unveil the intrinsic relation between the uncertainty and the risk tendency based on value at risk (VaR). Consequently, we propose two instantiations to model risk tendency, including an expert knowledge-based formulation embracing three essential properties and an adaptive learning method based on self-supervised reinforcement learning. We conduct extensive experiments on public datasets and show that the proposed framework outperforms state-of-the-art methods in practical settings.
Keyword: vision transformer
PanoViT: Vision Transformer for Room Layout Estimation from a Single Panoramic Image
Authors: Weichao Shen, Yuan Dong, Zonghao Chen, Zhengyi Zhao, Yang Gao, Zhu Liu
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
In this paper, we propose PanoViT, a panorama vision transformer to estimate the room layout from a single panoramic image. Compared to CNN models, our PanoViT is more proficient in learning global information from the panoramic image for the estimation of complex room layouts. Considering the difference between a perspective image and an equirectangular image, we design a novel recurrent position embedding and a patch sampling method for the processing of panoramic images. In addition to extracting global information, PanoViT also includes a frequency-domain edge enhancement module and a 3D loss to extract local geometric features in a panoramic image. Experimental results on several datasets demonstrate that our method outperforms state-of-the-art solutions in room layout prediction accuracy.
Keyword: multimodal
Do DALL-E and Flamingo Understand Each Other?
Authors: Hang Li, Jindong Gu, Rajat Koner, Sahand Sharifzadeh, Volker Tresp
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
A major goal of multimodal research is to improve machine understanding of images and text. Tasks include image captioning, text-to-image generation, and vision-language representation learning. So far, research has focused on the relationships between images and text. For example, captioning models attempt to understand the semantics of images which are then transformed into text. An important question is: which annotation reflects best a deep understanding of image content? Similarly, given a text, what is the best image that can present the semantics of the text? In this work, we argue that the best text or caption for a given image is the text which would generate the image which is the most similar to that image. Likewise, the best image for a given text is the image that results in the caption which is best aligned with the original text. To this end, we propose a unified framework that includes both a text-to-image generative model and an image-to-text generative model. Extensive experiments validate our approach.
Keyword: CLIP
When are Lemons Purple? The Concept Association Bias of CLIP
Abstract
Large-scale vision-language models such as CLIP have shown impressive performance on zero-shot image classification and image-to-text retrieval. However, such zero-shot performance of CLIP-based models does not realize in tasks that require a finer-grained correspondence between vision and language, such as Visual Question Answering (VQA). We investigate why this is the case, and report an interesting phenomenon of CLIP, which we call the Concept Association Bias (CAB), as a potential cause of the difficulty of applying CLIP to VQA and similar tasks. CAB is especially apparent when two concepts are present in the given image while a text prompt only contains a single concept. In such a case, we find that CLIP tends to treat input as a bag of concepts and attempts to fill in the other missing concept crossmodally, leading to an unexpected zero-shot prediction. For example, when asked for the color of a lemon in an image, CLIP predicts ``purple'' if the image contains a lemon and an eggplant. We demonstrate the Concept Association Bias of CLIP by showing that CLIP's zero-shot classification performance greatly suffers when there is a strong concept association between an object (e.g. lemon) and an attribute (e.g. its color). On the other hand, when the association between object and attribute is weak, we do not see this phenomenon. Furthermore, we show that CAB is significantly mitigated when we enable CLIP to learn deeper structure across image and text embeddings by adding an additional Transformer on top of CLIP and fine-tuning it on VQA. We find that across such fine-tuned variants of CLIP, the strength of CAB in a model predicts how well it performs on VQA.
Learning to Detect and Segment for Open Vocabulary Object Detection
Authors: Tao Wang, Nan Li
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Abstract
Open vocabulary object detection has been greatly advanced by the recent development of vision-language pretrained model, which helps recognize novel objects with only semantic categories. The prior works mainly focus on knowledge transferring to the object proposal classification and employ class-agnostic box and mask prediction. In this work, we propose CondHead, a principled dynamic network design to better generalize the box regression and mask segmentation for open vocabulary setting. The core idea is to conditionally parameterize the network heads on semantic embedding and thus the model is guided with class-specific knowledge to better detect novel categories. Specifically, CondHead is composed of two streams of network heads, the dynamically aggregated head and the dynamically generated head. The former is instantiated with a set of static heads that are conditionally aggregated, these heads are optimized as experts and are expected to learn sophisticated prediction. The latter is instantiated with dynamically generated parameters and encodes general class-specific information. With such a conditional design, the detection model is bridged by the semantic embedding to offer strongly generalizable class-wise box and mask prediction. Our method brings significant improvement to the state-of-the-art open vocabulary object detection methods with very minor overhead, e.g., it surpasses a RegionClip model by 3.0 detection AP on novel categories, with only 1.1% more computation.
Keyword: metric learning
There is no result
Keyword: image retrieval
There is no result
Keyword: self-supervised
Principled and Efficient Transfer Learning of Deep Models via Neural Collapse
Benchmark for Uncertainty & Robustness in Self-Supervised Learning
Adaptive Risk-Aware Bidding with Budget Constraint in Display Advertising
Keyword: vision transformer
PanoViT: Vision Transformer for Room Layout Estimation from a Single Panoramic Image
Keyword: multimodal
Do DALL-E and Flamingo Understand Each Other?
Keyword: CLIP
When are Lemons Purple? The Concept Association Bias of CLIP
Learning to Detect and Segment for Open Vocabulary Object Detection
Keyword: DALLE
There is no result