For service robots to become general-purpose in everyday household environments, they need not only a large library of primitive skills, but also the ability to quickly learn novel tasks specified by users. Fine-tuning neural networks on a variety of downstream tasks has been successful in many vision and language domains, but research is still limited on transfer learning between diverse long-horizon tasks. We propose that, compared to reinforcement learning for a new household activity from scratch, home robots can benefit from transferring the value and policy networks trained for similar tasks. We evaluate this idea in the BEHAVIOR simulation benchmark which includes a large number of household activities and a set of action primitives. For easy mapping between state spaces of different tasks, we provide a text-based representation and leverage language models to produce a common embedding space. The results show that the selection of similar source activities can be informed by the semantic similarity of state and goal descriptions with the target task. We further analyze the results and discuss ways to overcome the problem of catastrophic forgetting.
On robotics computer vision tasks, generating and annotating large amounts of data from real-world for the use of deep learning-based approaches is often difficult or even impossible. A common strategy for solving this problem is to apply simulation-to-reality (sim2real) approaches with the help of simulated scenes. While the majority of current robotics vision sim2real work focuses on image data, we present an industrial application case that uses sim2real transfer learning for point cloud data. We provide insights on how to generate and process synthetic point cloud data in order to achieve better performance when the learned model is transferred to real-world data. The issue of imbalanced learning is investigated using multiple strategies. A novel patch-based attention network is proposed additionally to tackle this problem.
To replace data augmentation, this paper proposed a method called SLAP to intensify experience to speed up machine learning and reduce the sample size. SLAP is a model-independent protocol/function to produce the same output given different transformation variants. SLAP improved the convergence speed of convolutional neural network learning by 83% in the experiments with Gomoku game states, with only one eighth of the sample size compared with data augmentation. In reinforcement learning for Gomoku, using AlphaGo Zero/AlphaZero algorithm with data augmentation as baseline, SLAP reduced the number of training samples by a factor of 8 and achieved similar winning rate against the same evaluator, but it was not yet evident that it could speed up reinforcement learning. The benefits should at least apply to domains that are invariant to symmetry or certain transformations. As future work, SLAP may aid more explainable learning and transfer learning for domains that are not invariant to symmetry, as a small step towards artificial general intelligence.
Self-supervised representation learning has been extremely successful in medical image analysis, as it requires no human annotations to provide transferable representations for downstream tasks. Recent self-supervised learning methods are dominated by noise-contrastive estimation (NCE, also known as contrastive learning), which aims to learn invariant visual representations by contrasting one homogeneous image pair with a large number of heterogeneous image pairs in each training step. Nonetheless, NCE-based approaches still suffer from one major problem that is one homogeneous pair is not enough to extract robust and invariant semantic information. Inspired by the archetypical triplet loss, we propose GraVIS, which is specifically optimized for learning self-supervised features from dermatology images, to group homogeneous dermatology images while separating heterogeneous ones. In addition, a hardness-aware attention is introduced and incorporated to address the importance of homogeneous image views with similar appearance instead of those dissimilar homogeneous ones. GraVIS significantly outperforms its transfer learning and self-supervised learning counterparts in both lesion segmentation and disease classification tasks, sometimes by 5 percents under extremely limited supervision. More importantly, when equipped with the pre-trained weights provided by GraVIS, a single model could achieve better results than winners that heavily rely on ensemble strategies in the well-known ISIC 2017 challenge.
Today's sign language recognition models require large training corpora of laboratory-like videos, whose collection involves an extensive workforce and financial resources. As a result, only a handful of such systems are publicly available, not to mention their limited localization capabilities for less-populated sign languages. Utilizing online text-to-video dictionaries, which inherently hold annotated data of various attributes and sign languages, and training models in a few-shot fashion hence poses a promising path for the democratization of this technology. In this work, we collect and open-source the UWB-SL-Wild few-shot dataset, the first of its kind training resource consisting of dictionary-scraped videos. This dataset represents the actual distribution and characteristics of available online sign language data. We select glosses that directly overlap with the already existing datasets WLASL100 and ASLLVD and share their class mappings to allow for transfer learning experiments. Apart from providing baseline results on a pose-based architecture, we introduce a novel approach to training sign language recognition models in a few-shot scenario, resulting in state-of-the-art results on ASLLVD-Skeleton and ASLLVD-Skeleton-20 datasets with top-1 accuracy of $30.97~\%$ and $95.45~\%$, respectively.
Consistent and holistic expression of software requirements is important for the success of software projects. In this study, we aim to enhance the efficiency of the software development processes by automatically identifying conflicting and duplicate software requirement specifications. We formulate the conflict and duplicate detection problem as a requirement pair classification task. We design a novel transformers-based architecture, SR-BERT, which incorporates Sentence-BERT and Bi-encoders for the conflict and duplicate identification task. Furthermore, we apply supervised multi-stage fine-tuning to the pre-trained transformer models. We test the performance of different transfer models using four different datasets. We find that sequentially trained and fine-tuned transformer models perform well across the datasets with SR-BERT achieving the best performance for larger datasets. We also explore the cross-domain performance of conflict detection models and adopt a rule-based filtering approach to validate the model classifications. Our analysis indicates that the sentence pair classification approach and the proposed transformer-based natural language processing strategies can contribute significantly to achieving automation in conflict and duplicate detection
Deep reinforcement learning (DRL) has been increasingly employed to handle the dynamic and complex resource management in network slicing. The deployment of DRL policies in real networks, however, is complicated by heterogeneous cell conditions. In this paper, we propose a novel transfer learning (TL) aided multi-agent deep reinforcement learning (MADRL) approach with inter-agent similarity analysis for inter-cell inter-slice resource partitioning. First, we design a coordinated MADRL method with information sharing to intelligently partition resource to slices and manage inter-cell interference. Second, we propose an integrated TL method to transfer the learned DRL policies among different local agents for accelerating the policy deployment. The method is composed of a new domain and task similarity measurement approach and a new knowledge transfer approach, which resolves the problem of from whom to transfer and how to transfer. We evaluated the proposed solution with extensive simulations in a system-level simulator and show that our approach outperforms the state-of-the-art solutions in terms of performance, convergence speed and sample efficiency. Moreover, by applying TL, we achieve an additional gain over 27% higher than the coordinate MADRL approach without TL.
We present a self-supervised, time-to-event (TTE) foundation model called MOTOR (Many Outcome Time Oriented Representations) which is pretrained on timestamped sequences of events in electronic health records (EHR) and health insurance claims. TTE models are used for estimating the probability distribution of the time until a specific event occurs, which is an important task in medical settings. TTE models provide many advantages over classification using fixed time horizons, including naturally handling censored observations, but are challenging to train with limited labeled data. MOTOR addresses this challenge by pretraining on up to 55M patient records (9B clinical events). We evaluate MOTOR's transfer learning performance on 19 tasks, across 3 patient databases (a private EHR system, MIMIC-IV, and Merative claims data). Task-specific models adapted from MOTOR improve time-dependent C statistics by 4.6% over state-of-the-art, improve label efficiency by up to 95% ,and are more robust to temporal distributional shifts. We further evaluate cross-site portability by adapting our MOTOR foundation model for six prediction tasks on the MIMIC-IV dataset, where it outperforms all baselines. MOTOR is the first foundation model for medical TTE predictions and we release a 143M parameter pretrained model for research use at [redacted URL].
Extraction of sentiment signals from news text, stock message boards, and business reports, for stock movement prediction, has been a rising field of interest in finance. Building upon past literature, the most recent works attempt to better capture sentiment from sentences with complex syntactic structures by introducing aspect-level sentiment classification (ASC). Despite the growing interest, however, fine-grained sentiment analysis has not been fully explored in non-English literature due to the shortage of annotated finance-specific data. Accordingly, it is necessary for non-English languages to leverage datasets and pre-trained language models (PLM) of different domains, languages, and tasks to best their performance. To facilitate finance-specific ASC research in the Korean language, we build KorFinASC, a Korean aspect-level sentiment classification dataset for finance consisting of 12,613 human-annotated samples, and explore methods of intermediate transfer learning. Our experiments indicate that past research has been ignorant towards the potentially wrong knowledge of financial entities encoded during the training phase, which has overestimated the predictive power of PLMs. In our work, we use the term "non-stationary knowledge'' to refer to information that was previously correct but is likely to change, and present "TGT-Masking'', a novel masking pattern to restrict PLMs from speculating knowledge of the kind. Finally, through a series of transfer learning with TGT-Masking applied we improve 22.63% of classification accuracy compared to standalone models on KorFinASC.
Current human activity recognition (HAR) techniques regard activity labels as integer class IDs without explicitly modeling the semantics of class labels. We observe that different activity names often have shared structures. For example, "open door" and "open fridge" both have "open" as the action; "kicking soccer ball" and "playing tennis ball" both have "ball" as the object. Such shared structures in label names can be translated to the similarity in sensory data and modeling common structures would help uncover knowledge across different activities, especially for activities with limited samples. In this paper, we propose SHARE, a HAR framework that takes into account shared structures of label names for different activities. To exploit the shared structures, SHARE comprises an encoder for extracting features from input sensory time series and a decoder for generating label names as a token sequence. We also propose three label augmentation techniques to help the model more effectively capture semantic structures across activities, including a basic token-level augmentation, and two enhanced embedding-level and sequence-level augmentations utilizing the capabilities of pre-trained models. SHARE outperforms state-of-the-art HAR models in extensive experiments on seven HAR benchmark datasets. We also evaluate in few-shot learning and label imbalance settings and observe even more significant performance gap.
Non-intrusive load monitoring (NILM) or energy disaggregation aims to extract the load profiles of individual consumer electronic appliances, given an aggregate load profile of the mains of a smart home. This work proposes a novel deep-learning and edge computing approach to solve the NILM problem and a few related problems as follows. 1) We build upon the reputed seq2-point convolutional neural network (CNN) model to come up with the proposed seq2-[3]-point CNN model to solve the (home) NILM problem and site-NILM problem (basically, NILM at a smaller scale). 2) We solve the related problem of appliance identification by building upon the state-of-the-art (pre-trained) 2D-CNN models, i.e., AlexNet, ResNet-18, and DenseNet-121, which are fine-tuned two custom datasets that consist of Wavelets and short-time Fourier transform (STFT)-based 2D electrical signatures of the appliances. 3) Finally, we do some basic qualitative inference about an individual appliance's health by comparing the power consumption of the same appliance across multiple homes. Low-frequency REDD dataset is used for all problems, except site-NILM where REFIT dataset has been used. As for the results, we achieve a maximum accuracy of 94.6\% for home-NILM, 81\% for site-NILM, and 88.9\% for appliance identification (with Resnet-based model).
Short-fiber-reinforced composites (SFRC) are high-performance engineering materials for lightweight structural applications in the automotive and electronics industries. Typically, SFRC structures are manufactured by injection molding, which induces heterogeneous microstructures, and the resulting nonlinear anisotropic behaviors are challenging to predict by conventional micromechanical analyses. In this work, we present a machine learning-based multiscale method by integrating injection molding-induced microstructures, material homogenization, and Deep Material Network (DMN) in the finite element simulation software LS-DYNA for structural analysis of SFRC. DMN is a physics-embedded machine learning model that learns the microscale material morphologies hidden in representative volume elements of composites through offline training. By coupling DMN with finite elements, we have developed a highly accurate and efficient data-driven approach, which predicts nonlinear behaviors of composite materials and structures at a computational speed orders-of-magnitude faster than the high-fidelity direct numerical simulation. To model industrial-scale SFRC products, transfer learning is utilized to generate a unified DMN database, which effectively captures the effects of injection molding-induced fiber orientations and volume fractions on the overall composite properties. Numerical examples are presented to demonstrate the promising performance of this LS-DYNA machine learning-based multiscale method for SFRC modeling.
Applying new computing paradigms like quantum computing to the field of machine learning has recently gained attention. However, as high-dimensional real-world applications are not yet feasible to be solved using purely quantum hardware, hybrid methods using both classical and quantum machine learning paradigms have been proposed. For instance, transfer learning methods have been shown to be successfully applicable to hybrid image classification tasks. Nevertheless, beneficial circuit architectures still need to be explored. Therefore, tracing the impact of the chosen circuit architecture and parameterization is crucial for the development of beneficially applicable hybrid methods. However, current methods include processes where both parts are trained concurrently, therefore not allowing for a strict separability of classical and quantum impact. Thus, those architectures might produce models that yield a superior prediction accuracy whilst employing the least possible quantum impact. To tackle this issue, we propose Sequential Quantum Enhanced Training (SEQUENT) an improved architecture and training process for the traceable application of quantum computing methods to hybrid machine learning. Furthermore, we provide formal evidence for the disadvantage of current methods and preliminary experimental results as a proof-of-concept for the applicability of SEQUENT.
With recent developments in digitization of clinical psychology, NLP research community has revolutionized the field of mental health detection on social media. Existing research in mental health analysis revolves around the cross-sectional studies to classify users' intent on social media. For in-depth analysis, we investigate existing classifiers to solve the problem of causal categorization which suggests the inefficiency of learning based methods due to limited training samples. To handle this challenge, we use transformer models and demonstrate the efficacy of a pre-trained transfer learning on "CAMS" dataset. The experimental result improves the accuracy and depicts the importance of identifying cause-and-effect relationships in the underlying text.
Advancements in Text-to-Image synthesis over recent years have focused more on improving the quality of generated samples using datasets with descriptive prompts. However, real-world image-caption pairs present in domains such as news data do not use simple and directly descriptive captions. With captions containing information on both the image content and underlying contextual cues, they become abstractive in nature. In this paper, we launch ANNA, an Abstractive News captioNs dAtaset extracted from online news articles in a variety of different contexts. We explore the capabilities of current Text-to-Image synthesis models to generate news domain-specific images using abstractive captions by benchmarking them on ANNA, in both standard training and transfer learning settings. The generated images are judged on the basis of contextual relevance, visual quality, and perceptual similarity to ground-truth image-caption pairs. Through our experiments, we show that techniques such as transfer learning achieve limited success in understanding abstractive captions but still fail to consistently learn the relationships between content and context features. The Dataset is available at https://github.com/aashish2000/ANNA .
We introduce multi-head neural networks (MH-NNs) to physics-informed machine learning, which is a type of neural networks (NNs) with all nonlinear hidden layers as the body and multiple linear output layers as multi-head. Hence, we construct multi-head physics-informed neural networks (MH-PINNs) as a potent tool for multi-task learning (MTL), generative modeling, and few-shot learning for diverse problems in scientific machine learning (SciML). MH-PINNs connect multiple functions/tasks via a shared body as the basis functions as well as a shared distribution for the head. The former is accomplished by solving multiple tasks with MH-PINNs with each head independently corresponding to each task, while the latter by employing normalizing flows (NFs) for density estimate and generative modeling. To this end, our method is a two-stage method, and both stages can be tackled with standard deep learning tools of NNs, enabling easy implementation in practice. MH-PINNs can be used for various purposes, such as approximating stochastic processes, solving multiple tasks synergistically, providing informative prior knowledge for downstream few-shot learning tasks such as meta-learning and transfer learning, learning representative basis functions, and uncertainty quantification. We demonstrate the effectiveness of MH-PINNs in five benchmarks, investigating also the possibility of synergistic learning in regression analysis. We name the open-source code "Lernaean Hydra" (L-HYDRA), since this mythical creature possessed many heads for performing important multiple tasks, as in the proposed method.
Breast cancer is the second most common cancer among women worldwide. Diagnosis of breast cancer by the pathologists is a time-consuming procedure and subjective. Computer aided diagnosis frameworks are utilized to relieve pathologist workload by classifying the data automatically, in which deep convolutional neural networks (CNNs) are effective solutions. The features extracted from activation layer of pre-trained CNNs are called deep convolutional activation features (DeCAF). In this paper, we have analyzed that all DeCAF features are not necessarily led to a higher accuracy in the classification task and dimension reduction plays an important role. Therefore, different dimension reduction methods are applied to achieve an effective combination of features by capturing the essence of DeCAF features. To this purpose, we have proposed reduced deep convolutional activation features (R-DeCAF). In this framework, pre-trained CNNs such as AlexNet, VGG-16 and VGG-19 are utilized in transfer learning mode as feature extractors. DeCAF features are extracted from the first fully connected layer of the mentioned CNNs and support vector machine has been used for binary classification. Among linear and nonlinear dimensionality reduction algorithms, linear approaches such as principal component analysis (PCA) represent a better combination among deep features and lead to a higher accuracy in the classification task using small number of features considering specific amount of cumulative explained variance (CEV) of features. The proposed method is validated using experimental BreakHis dataset. Comprehensive results show improvement in the classification accuracy up to 4.3% with less computational time. Best achieved accuracy is 91.13% for 400x data with feature vector size (FVS) of 23 and CEV equals to 0.15 using pre-trained AlexNet as feature extractor and PCA as feature reduction algorithm.
This paper proposes a pre-trained neural network for handling event camera data. Our model is a self-supervised learning framework, and uses paired event camera data and natural RGB images for training. Our method contains three modules connected in a sequence: i) a family of event data augmentations, generating meaningful event images for self-supervised training; ii) a conditional masking strategy to sample informative event patches from event images, encouraging our model to capture the spatial layout of a scene and accelerating training; iii) a contrastive learning approach, enforcing the similarity of embeddings between matching event images, and between paired event and RGB images. An embedding projection loss is proposed to avoid the model collapse when enforcing the event image embedding similarities. A probability distribution alignment loss is proposed to encourage the event image to be consistent with its paired RGB image in the feature space. Transfer learning performance on downstream tasks shows the superiority of our method over state-of-the-art methods. For example, we achieve top-1 accuracy at 64.83% on the N-ImageNet dataset.
Due to its probabilistic nature, fault prognostics is a prime example of a use case for deep learning utilizing big data. However, the low availability of such data sets combined with the high effort of fitting, parameterizing and evaluating complex learning algorithms to the heterogenous and dynamic settings typical for industrial applications oftentimes prevents the practical application of this approach. Automatic adaptation to new or dynamically changing fault prognostics scenarios can be achieved using transfer learning or continual learning methods. In this paper, a first survey of such approaches is carried out, aiming at establishing best practices for future research in this field. It is shown that the field is lacking common benchmarks to robustly compare results and facilitate scientific progress. Therefore, the data sets utilized in these publications are surveyed as well in order to identify suitable candidates for such benchmark scenarios.
Dynamic Magnetic Resonance Imaging (dMRI) is widely used to assess various cardiac conditions such as cardiac motion and blood flow. To accelerate MR acquisition, techniques such as undersampling and Simultaneous Multi-Slice (SMS) are often used. Special reconstruction algorithms are needed to reconstruct multiple SMS image slices from the entangled information. Deep learning (DL)-based methods have shown promising results for single-slice MR reconstruction, but the addition of SMS acceleration raises unique challenges due to the composite k-space signals and the resulting images with strong inter-slice artifacts. Furthermore, many dMRI applications lack sufficient data for training reconstruction neural networks. In this study, we propose a novel DL-based framework for dynamic SMS reconstruction. Our main contributions are 1) a combination of data transformation steps and network design that effectively leverages the unique characteristics of undersampled dynamic SMS data, and 2) an MR physics-guided transfer learning strategy that addresses the data scarcity issue. Thorough comparisons with multiple baseline methods illustrate the strengths of our proposed methods.