In representation learning, regression has traditionally received less attention than classification. Directly applying representation learning techniques designed for classification to regression often results in fragmented representations in the latent space, yielding sub-optimal performance. In this paper, we argue that the potential of contrastive learning for regression has been overshadowed due to the neglect of two crucial aspects: ordinality-awareness and hardness. To address these challenges, we advocate "mixup your own contrastive pairs for supervised contrastive regression", instead of relying solely on real/augmented samples. Specifically, we propose Supervised Contrastive Learning for Regression with Mixup (SupReMix). It takes anchor-inclusive mixtures (mixup of the anchor and a distinct negative sample) as hard negative pairs and anchor-exclusive mixtures (mixup of two distinct negative samples) as hard positive pairs at the embedding level. This strategy formulates harder contrastive pairs by integrating richer ordinal information. Through extensive experiments on six regression datasets including 2D images, volumetric images, text, tabular data, and time-series signals, coupled with theoretical analysis, we demonstrate that SupReMix pre-training fosters continuous ordered representations of regression data, resulting in significant improvement in regression performance. Furthermore, SupReMix is superior to other approaches in a range of regression challenges including transfer learning, imbalanced training data, and scenarios with fewer training samples.
Bayesian optimization (BO) is a popular black-box function optimization method, which makes sequential decisions based on a Bayesian model, typically a Gaussian process (GP), of the function. To ensure the quality of the model, transfer learning approaches have been developed to automatically design GP priors by learning from observations on "training" functions. These training functions are typically required to have the same domain as the "test" function (black-box function to be optimized). In this paper, we introduce MPHD, a model pre-training method on heterogeneous domains, which uses a neural net mapping from domain-specific contexts to specifications of hierarchical GPs. MPHD can be seamlessly integrated with BO to transfer knowledge across heterogeneous search spaces. Our theoretical and empirical results demonstrate the validity of MPHD and its superior performance on challenging black-box function optimization tasks.
This study explored the application of CNN-Transfer Learning for nondestructive chicken egg fertility detection for precision poultry hatchery practices. Four models, VGG16, ResNet50, InceptionNet, and MobileNet, were trained and evaluated on a dataset (200 single egg images) using augmented images (rotation, flip, scale, translation, and reflection). Although the training results demonstrated that all models achieved high accuracy, indicating their ability to accurately learn and classify chicken eggs' fertility state, when evaluated on the testing set, variations in accuracy and performance were observed. InceptionNet exhibited the best overall performance, accurately classifying fertile and non-fertile eggs. It demonstrated excellent performance in both training and testing sets in all parameters of the evaluation metrics. In testing set, it achieved an accuracy of 0.98, a sensitivity of 1 for detecting fertile eggs, and a specificity of 0.96 for identifying non-fertile eggs. The higher performance is attributed to its unique architecture efficiently capturing features at different scales leading to improved accuracy and robustness. Further optimization and fine-tuning of the models might necessary to address the limitations in accurately detecting fertile and non-fertile eggs in case of other models. This study highlighted the potential of CNN-Transfer Learning for nondestructive fertility detection and emphasizes the need for further research to enhance the models' capabilities and ensure accurate classification.
Continual Learning methods are designed to learn new tasks without erasing previous knowledge. However, Continual Learning often requires massive computational power and storage capacity for satisfactory performance. In this paper, we propose a resource-efficient continual learning method called the Elastic Expansion Network (E2Net). Leveraging core subnet distillation and precise replay sample selection, E2Net achieves superior average accuracy and diminished forgetting within the same computational and storage constraints, all while minimizing processing time. In E2Net, we propose Representative Network Distillation to identify the representative core subnet by assessing parameter quantity and output similarity with the working network, distilling analogous subnets within the working network to mitigate reliance on rehearsal buffers and facilitating knowledge transfer across previous tasks. To enhance storage resource utilization, we then propose Subnet Constraint Experience Replay to optimize rehearsal efficiency through a sample storage strategy based on the structures of representative networks. Extensive experiments conducted predominantly on cloud environments with diverse datasets and also spanning the edge environment demonstrate that E2Net consistently outperforms state-of-the-art methods. In addition, our method outperforms competitors in terms of both storage and computational requirements.
Recent works in end-to-end speech-to-text translation (ST) have proposed multi-tasking methods with soft parameter sharing which leverage machine translation (MT) data via secondary encoders that map text inputs to an eventual cross-modal representation. In this work, we instead propose a ST/MT multi-tasking framework with hard parameter sharing in which all model parameters are shared cross-modally. Our method reduces the speech-text modality gap via a pre-processing stage which converts speech and text inputs into two discrete token sequences of similar length -- this allows models to indiscriminately process both modalities simply using a joint vocabulary. With experiments on MuST-C, we demonstrate that our multi-tasking framework improves attentional encoder-decoder, Connectionist Temporal Classification (CTC), transducer, and joint CTC/attention models by an average of +0.5 BLEU without any external MT data. Further, we show that this framework incorporates external MT data, yielding +0.8 BLEU, and also improves transfer learning from pre-trained textual models, yielding +1.8 BLEU.
Precise answers are extracted from a text for a given input question in a question answering system. Marathi question answering system is created in recent studies by using ontology, rule base and machine learning based approaches. Recently transformer models and transfer learning approaches are used to solve question answering challenges. In this paper we investigate different transformer models for creating a reading comprehension-based Marathi question answering system. We have experimented on different pretrained Marathi language multilingual and monolingual models like Multilingual Representations for Indian Languages (MuRIL), MahaBERT, Indic Bidirectional Encoder Representations from Transformers (IndicBERT) and fine-tuned it on a Marathi reading comprehension-based data set. We got the best accuracy in a MuRIL multilingual model with an EM score of 0.64 and F1 score of 0.74 by fine tuning the model on the Marathi dataset.
Unsupervised domain adaptation aims to transfer knowledge from a fully-labeled source domain to an unlabeled target domain. However, in real-world scenarios, providing abundant labeled data even in the source domain can be infeasible due to the difficulty and high expense of annotation. To address this issue, recent works consider the Few-shot Unsupervised Domain Adaptation (FUDA) where only a few source samples are labeled, and conduct knowledge transfer via self-supervised learning methods. Yet existing methods generally overlook that the sparse label setting hinders learning reliable source knowledge for transfer. Additionally, the learning difficulty difference in target samples is different but ignored, leaving hard target samples poorly classified. To tackle both deficiencies, in this paper, we propose a novel Confidence-based Visual Dispersal Transfer learning method (C-VisDiT) for FUDA. Specifically, C-VisDiT consists of a cross-domain visual dispersal strategy that transfers only high-confidence source knowledge for model adaptation and an intra-domain visual dispersal strategy that guides the learning of hard target samples with easy ones. We conduct extensive experiments on Office-31, Office-Home, VisDA-C, and DomainNet benchmark datasets and the results demonstrate that the proposed C-VisDiT significantly outperforms state-of-the-art FUDA methods. Our code is available at https://github.com/Bostoncake/C-VisDiT.
Multimodal transfer learning aims to transform pretrained representations of diverse modalities into a common domain space for effective multimodal fusion. However, conventional systems are typically built on the assumption that all modalities exist, and the lack of modalities always leads to poor inference performance. Furthermore, extracting pretrained embeddings for all modalities is computationally inefficient for inference. In this work, to achieve high efficiency-performance multimodal transfer learning, we propose VideoAdviser, a video knowledge distillation method to transfer multimodal knowledge of video-enhanced prompts from a multimodal fundamental model (teacher) to a specific modal fundamental model (student). With an intuition that the best learning performance comes with professional advisers and smart students, we use a CLIP-based teacher model to provide expressive multimodal knowledge supervision signals to a RoBERTa-based student model via optimizing a step-distillation objective loss -- first step: the teacher distills multimodal knowledge of video-enhanced prompts from classification logits to a regression logit -- second step: the multimodal knowledge is distilled from the regression logit of the teacher to the student. We evaluate our method in two challenging multimodal tasks: video-level sentiment analysis (MOSI and MOSEI datasets) and audio-visual retrieval (VEGAS dataset). The student (requiring only the text modality as input) achieves an MAE score improvement of up to 12.3% for MOSI and MOSEI. Our method further enhances the state-of-the-art method by 3.4% mAP score for VEGAS without additional computations for inference. These results suggest the strengths of our method for achieving high efficiency-performance multimodal transfer learning.
By exploiting complementary sensor information, radar and camera fusion systems have the potential to provide a highly robust and reliable perception system for advanced driver assistance systems and automated driving functions. Recent advances in camera-based object detection offer new radar-camera fusion possibilities with bird's eye view feature maps. In this work, we propose a novel and flexible fusion network and evaluate its performance on two datasets: nuScenes and View-of-Delft. Our experiments reveal that while the camera branch needs large and diverse training data, the radar branch benefits more from a high-performance radar. Using transfer learning, we improve the camera's performance on the smaller dataset. Our results further demonstrate that the radar-camera fusion approach significantly outperforms the camera-only and radar-only baselines.
We present a holistic approach for high resolution image classification that won second place in the ICCV/CVPPA2023 Deep Nutrient Deficiency Challenge. The approach consists of a full pipeline of: 1) data distribution analysis to check potential domain shift, 2) backbone selection for a strong baseline model that scales up for high resolution input, 3) transfer learning that utilizes published pretrained models and continuous fine-tuning on small sub-datasets, 4) data augmentation for the diversity of training data and to prevent overfitting, 5) test-time augmentation to improve the prediction's robustness, and 6) "data soups" that conducts cross-fold model prediction average for smoothened final test results.
This paper which is part of the New Faculty Highlights Invited Speaker Program of AAAI'23, serves as a comprehensive survey of my research in transfer learning by utilizing embedding spaces. The work reviewed in this paper specifically revolves around the inherent challenges associated with continual learning and limited availability of labeled data. By providing an overview of my past and ongoing contributions, this paper aims to present a holistic understanding of my research, paving the way for future explorations and advancements in the field. My research delves into the various settings of transfer learning, including, few-shot learning, zero-shot learning, continual learning, domain adaptation, and distributed learning. I hope this survey provides a forward-looking perspective for researchers who would like to focus on similar research directions.
In the extensive recommender systems literature, novelty and diversity have been identified as key properties of useful recommendations. However, these properties have received limited attention in the specific sub-field of research paper recommender systems. In this work, we argue for the importance of offering novel and diverse research paper recommendations to scientists. This approach aims to reduce siloed reading, break down filter bubbles, and promote interdisciplinary research. We propose a novel framework for evaluating the novelty and diversity of research paper recommendations that leverages methods from network analysis and natural language processing. Using this framework, we show that the choice of representational method within a larger research paper recommendation system can have a measurable impact on the nature of downstream recommendations, specifically on their novelty and diversity. We highlight a novel paper embedding method, which we demonstrate offers more innovative and diverse recommendations without sacrificing precision, compared to other state-of-the-art baselines.
This study aims to evaluate the performance of deep learning models in predicting $\geq$M-class solar flares with a prediction window of 24 hours, using hourly sampled full-disk line-of-sight (LoS) magnetogram images, particularly focusing on the often overlooked flare events corresponding to the near-limb regions (beyond $\pm$70$^{\circ}$ of the solar disk). We trained three well-known deep learning architectures--AlexNet, VGG16, and ResNet34 using transfer learning and compared and evaluated the overall performance of our models using true skill statistics (TSS) and Heidke skill score (HSS) and computed recall scores to understand the prediction sensitivity in central and near-limb regions for both X- and M-class flares. The following points summarize the key findings of our study: (1) The highest overall performance was observed with the AlexNet-based model, which achieved an average TSS$\sim$0.53 and HSS$\sim$0.37; (2) Further, a spatial analysis of recall scores disclosed that for the near-limb events, the VGG16- and ResNet34-based models exhibited superior prediction sensitivity. The best results, however, were seen with the ResNet34-based model for the near-limb flares, where the average recall was approximately 0.59 (the recall for X- and M-class was 0.81 and 0.56 respectively) and (3) Our research findings demonstrate that our models are capable of discerning complex spatial patterns from full-disk magnetograms and exhibit skill in predicting solar flares, even in the vicinity of near-limb regions. This ability holds substantial importance for operational flare forecasting systems.
3D perception based on the representations learned from multi-camera bird's-eye-view (BEV) is trending as cameras are cost-effective for mass production in autonomous driving industry. However, there exists a distinct performance gap between multi-camera BEV and LiDAR based 3D object detection. One key reason is that LiDAR captures accurate depth and other geometry measurements, while it is notoriously challenging to infer such 3D information from merely image input. In this work, we propose to boost the representation learning of a multi-camera BEV based student detector by training it to imitate the features of a well-trained LiDAR based teacher detector. We propose effective balancing strategy to enforce the student to focus on learning the crucial features from the teacher, and generalize knowledge transfer to multi-scale layers with temporal fusion. We conduct extensive evaluations on multiple representative models of multi-camera BEV. Experiments reveal that our approach renders significant improvement over the student models, leading to the state-of-the-art performance on the popular benchmark nuScenes.
Image acquisition conditions and environments can significantly affect high-level tasks in computer vision, and the performance of most computer vision algorithms will be limited when trained on distortion-free datasets. Even with updates in hardware such as sensors and deep learning methods, it will still not work in the face of variable conditions in real-world applications. In this paper, we apply the object detector YOLOv7 to detect distorted images from the dataset CDCOCO. Through carefully designed optimizations including data enhancement, detection box ensemble, denoiser ensemble, super-resolution models, and transfer learning, our model achieves excellent performance on the CDCOCO test set. Our denoising detection model can denoise and repair distorted images, making the model useful in a variety of real-world scenarios and environments.
This study aims to explore efficient tuning methods for the screenshot captioning task. Recently, image captioning has seen significant advancements, but research in captioning tasks for mobile screens remains relatively scarce. Current datasets and use cases describing user behaviors within product screenshots are notably limited. Consequently, we sought to fine-tune pre-existing models for the screenshot captioning task. However, fine-tuning large pre-trained models can be resource-intensive, requiring considerable time, computational power, and storage due to the vast number of parameters in image captioning models. To tackle this challenge, this study proposes a combination of adapter methods, which necessitates tuning only the additional modules on the model. These methods are originally designed for vision or language tasks, and our intention is to apply them to address similar challenges in screenshot captioning. By freezing the parameters of the image caption models and training only the weights associated with the methods, performance comparable to fine-tuning the entire model can be achieved, while significantly reducing the number of parameters. This study represents the first comprehensive investigation into the effectiveness of combining adapters within the context of the screenshot captioning task. Through our experiments and analyses, this study aims to provide valuable insights into the application of adapters in vision-language models and contribute to the development of efficient tuning techniques for the screenshot captioning task. Our study is available at https://github.com/RainYuGG/BLIP-Adapter
Monocular depth estimation is challenging due to its inherent ambiguity and ill-posed nature, yet it is quite important to many applications. While recent works achieve limited accuracy by designing increasingly complicated networks to extract features with limited spatial geometric cues from a single RGB image, we intend to introduce spatial cues by training a teacher network that leverages left-right image pairs as inputs and transferring the learned 3D geometry-aware knowledge to the monocular student network. Specifically, we present a novel knowledge distillation framework, named ADU-Depth, with the goal of leveraging the well-trained teacher network to guide the learning of the student network, thus boosting the precise depth estimation with the help of extra spatial scene information. To enable domain adaptation and ensure effective and smooth knowledge transfer from teacher to student, we apply both attention-adapted feature distillation and focal-depth-adapted response distillation in the training stage. In addition, we explicitly model the uncertainty of depth estimation to guide distillation in both feature space and result space to better produce 3D-aware knowledge from monocular observations and thus enhance the learning for hard-to-predict image regions. Our extensive experiments on the real depth estimation datasets KITTI and DrivingStereo demonstrate the effectiveness of the proposed method, which ranked 1st on the challenging KITTI online benchmark.
Tracking using bio-inspired event cameras has drawn more and more attention in recent years. Existing works either utilize aligned RGB and event data for accurate tracking or directly learn an event-based tracker. The first category needs more cost for inference and the second one may be easily influenced by noisy events or sparse spatial resolution. In this paper, we propose a novel hierarchical knowledge distillation framework that can fully utilize multi-modal / multi-view information during training to facilitate knowledge transfer, enabling us to achieve high-speed and low-latency visual tracking during testing by using only event signals. Specifically, a teacher Transformer-based multi-modal tracking framework is first trained by feeding the RGB frame and event stream simultaneously. Then, we design a new hierarchical knowledge distillation strategy which includes pairwise similarity, feature representation, and response maps-based knowledge distillation to guide the learning of the student Transformer network. Moreover, since existing event-based tracking datasets are all low-resolution ($346 \times 260$), we propose the first large-scale high-resolution ($1280 \times 720$) dataset named EventVOT. It contains 1141 videos and covers a wide range of categories such as pedestrians, vehicles, UAVs, ping pongs, etc. Extensive experiments on both low-resolution (FE240hz, VisEvent, COESOT), and our newly proposed high-resolution EventVOT dataset fully validated the effectiveness of our proposed method. The dataset, evaluation toolkit, and source code are available on \url{https://github.com/Event-AHU/EventVOT_Benchmark}
Image colorization is the process of colorizing grayscale images or recoloring an already-color image. This image manipulation can be used for grayscale satellite, medical and historical images making them more expressive. With the help of the increasing computation power of deep learning techniques, the colorization algorithms results are becoming more realistic in such a way that human eyes cannot differentiate between natural and colorized images. However, this poses a potential security concern, as forged or illegally manipulated images can be used illegally. There is a growing need for effective detection methods to distinguish between natural color and computer-colorized images. This paper presents a novel approach that combines the advantages of transfer and ensemble learning approaches to help reduce training time and resource requirements while proposing a model to classify natural color and computer-colorized images. The proposed model uses pre-trained branches VGG16 and Resnet50, along with Mobile Net v2 or Efficientnet feature vectors. The proposed model showed promising results, with accuracy ranging from 94.55% to 99.13% and very low Half Total Error Rate values. The proposed model outperformed existing state-of-the-art models regarding classification performance and generalization capabilities.
Interpreting regulatory documents or building codes into computer-processable formats is essential for the intelligent design and construction of buildings and infrastructures. Although automated rule interpretation (ARI) methods have been investigated for years, most of them highly depend on the early and manual filtering of interpretable clauses from a building code. While few of them considered machine interpretability, which represents the potential to be transformed into a computer-processable format, from both clause- and document-level. Therefore, this research aims to propose a novel approach to automatically evaluate and enhance the machine interpretability of single clause and building codes. First, a few categories are introduced to classify each clause in a building code considering the requirements for rule interpretation, and a dataset is developed for model training. Then, an efficient text classification model is developed based on a pretrained domain-specific language model and transfer learning techniques. Finally, a quantitative evaluation method is proposed to assess the overall interpretability of building codes. Experiments show that the proposed text classification algorithm outperforms the existing CNN- or RNN-based methods, improving the F1-score from 72.16% to 93.60%. It is also illustrated that the proposed classification method can enhance downstream ARI methods with an improvement of 4%. Furthermore, analyzing the results of more than 150 building codes in China showed that their average interpretability is 34.40%, which implies that it is still hard to fully transform the entire regulatory document into computer-processable formats. It is also argued that the interpretability of building codes should be further improved both from the human side and the machine side.