Few-shot Learning Tasks Research Articles

Graph neural processes for molecules: an evaluation on docking scores and strategies to improve generalization

Neural processes (NPs) are models for meta-learning which output uncertainty estimates. So far, most studies of NPs have focused on low-dimensional datasets of highly-correlated tasks. While these homogeneous datasets are useful for benchmarking, they may not be representative of realistic transfer learning. In particular, applications in scientific research may prove especially challenging due to the potential novelty of meta-testing tasks. Molecular property prediction is one such research area that is characterized by sparse datasets of many functions on a shared molecular space. In this paper, we study the application of graph NPs to molecular property prediction with DOCKSTRING, a diverse dataset of docking scores. Graph NPs show competitive performance in few-shot learning tasks relative to supervised learning baselines common in chemoinformatics, as well as alternative techniques for transfer learning and meta-learning. In order to increase meta-generalization to divergent test functions, we propose fine-tuning strategies that adapt the parameters of NPs. We find that adaptation can substantially increase NPs' regression performance while maintaining good calibration of uncertainty estimates. Finally, we present a Bayesian optimization experiment which showcases the potential advantages of NPs over Gaussian processes in iterative screening. Overall, our results suggest that NPs on molecular graphs hold great potential for molecular property prediction in the low-data setting.Scientific contributionNeural processes are a family of meta-learning algorithms which deal with data scarcity by transferring information across tasks and making probabilistic predictions. We evaluate their performance on regression and optimization molecular tasks using docking scores, finding them to outperform classical single-task and transfer-learning models. We examine the issue of generalization to divergent test tasks, which is a general concern of meta-learning algorithms in science, and propose strategies to alleviate it.

A Suite of Foundation Models Captures the Contextual Interplay Between Codons.

In the canonical genetic code, many amino acids are assigned more than one codon. Work by us and others has shown that the choice of these synonymous codon is not random, and carries regulatory and functional consequences. Existing protein foundation models ignore this context-dependent role of coding sequence in shaping the protein landscape of the cell. To address this gap, we introduce cdsFM, a suite of codon-resolution large language models, including both EnCodon and DeCodon models, with up to 1B parameters. Pre-trained on 60 million protein-coding sequences from more than 5,000 species, our models effectively learn the relationship between codons and amino acids, recapitualing the overall structure of the genetic code. In addition to outperforming state-of-the-art genomic foundation models in a variety of zero-shot and few-shot learning tasks, the larger pre-trained models were superior in predicting the choice of synonymous codons. To systematically assess the impact of synonymous codon choices on protein expression and our models' ability to capture these effects, we generated a large dataset measuring overall and surface expression levels of three proteins as a function of changes in their synonymous codons. We showed that our EnCodon models could be readily fine-tuned to predict the contextual consequences of synonymous codon choices. Armed with this knowledge, we applied EnCodon to existing clinical datasets of synonymous variants, and we identified a large number of synonymous codons that are likely pathogenic, several of which we experimentally confirmed in a cell-based model. Together, our findings establish the cdsFM suite as a powerful tool for decoding the complex functional grammar underlying the choice of synonymous codons.

Unified View Empirical Study for Large Pretrained Model on Cross-Domain Few-Shot Learning

The challenge of cross-domain few-shot learning (CD-FSL) stems from the substantial distribution disparities between target and source domain images, necessitating a model with robust generalization capabilities. In this work, we posit that large-scale pretrained models are pivotal in addressing the CD-FSL task owing to their exceptional representational and generalization prowess. To our knowledge, no existing research comprehensively investigates the utility of large-scale pretrained models in the CD-FSL context. Addressing this gap, our study presents an exhaustive empirical assessment of the Contrastive Language–Image Pre-Training model within the CD-FSL task. We undertake a comparison spanning six dimensions: base model, transfer module, classifier, loss, data augmentation, and training schedule. Furthermore, we establish a straightforward baseline model, E-base, based on our empirical analysis, underscoring the importance of our investigation. Experimental results substantiate the efficacy of our model, yielding a mean gain of 1.2% in 5-way 5-shot evaluations on the BSCD dataset.

Few-Shot Learning in Wi-Fi-Based Indoor Positioning.

This paper explores the use of few-shot learning in Wi-Fi-based indoor positioning, utilizing convolutional neural networks (CNNs) combined with meta-learning techniques to enhance the accuracy and efficiency of positioning systems. The focus is on addressing the challenge of limited labeled data, a prevalent issue in extensive indoor environments. The study explores various scenarios, comparing the performance of the base CNN and meta-learning models. The meta-learning approach involves few-shot learning tasks, such as three-way N-shot, five-way N-shot, etc., to enhance the model's ability to generalize from limited data. The experiments were conducted across various scenarios, evaluating the performance of the models with different numbers of samples per class (K) after filtering by cosine similarity (FCS) during both the stages of data preprocessing and meta-learning. The scenarios included both base classes and novel classes, with and without meta-learning. The results indicated that the base CNN model achieved varying accuracy levels depending on the scenario and the number of samples per class retained after FCS. Meta-learning performed acceptably in scenarios with fewer samples, which are the distinct datasets pertaining to novel classes. With 20 samples per class, the base CNN achieved an accuracy of 0.80 during the pre-training stage, while meta-learning (three-way one-shot) achieved an accuracy of 0.78 on a new small dataset with novel classes.

Less is more: A closer look at semantic-based few-shot learning

Few-shot Learning (FSL) aims to learn and distinguish new categories from a scant number of available samples, presenting a significant challenge in the realm of deep learning. Recent researchers have sought to leverage the additional semantic or linguistic information of scarce categories with a pre-trained language model to facilitate learning, thus partially alleviating the problem of insufficient supervision signals. Nonetheless, the full potential of the semantic information and pre-trained language model have been underestimated in the few-shot learning till now, resulting in limited performance enhancements. To address this, we propose a straightforward and efficacious framework for few-shot learning tasks, specifically designed to exploit the semantic information and language model. Specifically, we explicitly harness the zero-shot capability of the pre-trained language model with learnable prompts. And we directly add the visual feature with the textual feature for inference without the intricate designed fusion modules as in prior studies. Additionally, we apply the self-ensemble and distillation to further enhance performance. Extensive experiments conducted across four widely used few-shot datasets demonstrate that our simple framework achieves impressive results. Particularly noteworthy is its outstanding performance in the 1-shot learning task, surpassing the current state-of-the-art by an average of 3.3% in classification accuracy. Our code will be available at https://github.com/zhouchunpong/SimpleFewShot.

Few-shot learning with representative global prototype

Few-shot learning is often challenged by low generalization performance due to the model is mostly learned with the base classes only. To mitigate the above issues, a few-shot learning method with representative global prototype is proposed in this paper. Specifically, to enhance generalization to novel class, we propose a strategy for jointly training base and novel classes. This process produces prototypes characterizing the class information called representative global prototypes. Additionally, to avoid the problem of data imbalance and prototype bias caused by newly added categories of sparse samples, a novel sample synthesis method is proposed for augmenting more representative samples of novel class. Finally, representative samples and non-representative samples with high uncertainty are selected to enhance the representational and discriminative abilities of the global prototype. Intensive experiments have been conducted on two popular benchmark datasets, and the experimental results show that this method significantly improves the classification ability of few-shot learning tasks and achieves state-of-the-art performance.

Multi-scale task-aware structure graph modeling for few-shot image recognition

The Few-shot image recognition attempts to recognize images from a novel class with only a limited number of labeled training images, which is a few-shot learning (FSL) task. FSL is very challenging. Limited labeled training samples cannot adequately represent the distribution of classes, and the base and novel classes in the training and testing stages do not intersect and have different distributions, leading to a domain shift problem in generalizing the learned model to the novel class dataset. In this paper, we propose multi-scale task-aware structure graph modeling for few-shot image recognition. We train a meta-filter learner to generate task-aware local structure filters for each scale and adaptively capture the local structures at each scale. Moreover, we introduce a novel multi-scale graph attention network (MGAT) module to model the multi-scale local structures of an image, fully exploring the correlations between different local structures at all scales of the image. Finally, we leverage the attention mechanism of graph attention network to achieve information aggregation and propagation, aiming to obtain more representative and discriminative local structure features that integrate both local and global information. We conducted comprehensive experiments on four benchmark datasets widely adopted in FSL tasks. The experimental results demonstrate that the MTSGM obtains state-of-the-art performance, which validates the effectiveness of MTSGM.

Reshaping Bioacoustics Event Detection: Leveraging Few-Shot Learning (FSL) with Transductive Inference and Data Augmentation.

Bioacoustic event detection is a demanding endeavor involving recognizing and classifying the sounds animals make in their natural habitats. Traditional supervised learning requires a large amount of labeled data, which are hard to come by in bioacoustics. This paper presents a few-shot learning (FSL) method incorporating transductive inference and data augmentation to address the issues of too few labeled events and small volumes of recordings. Here, transductive inference iteratively alters class prototypes and feature extractors to seize essential patterns, whereas data augmentation applies SpecAugment on Mel spectrogram features to augment training data. The proposed approach is evaluated by using the Detecting and Classifying Acoustic Scenes and Events (DCASE) 2022 and 2021 datasets. Extensive experimental results demonstrate that all components of the proposed method achieve significant F-score improvements of 27% and 10%, for the DCASE-2022 and DCASE-2021 datasets, respectively, compared to recent advanced approaches. Moreover, our method is helpful in FSL tasks because it effectively adapts to sounds from various animal species, recordings, and durations.

A fine-grained self-adapting prompt learning approach for few-shot learning with pre-trained language models

Pre-trained language models have demonstrated remarkable performance in few-shot learning through the emergence of “prompt-based learning” methods, where the performance of these tasks highly rely on the quality of prompts. Existing prompt learning methods typically customize a single prompt to each few-shot learning task and all the examples in the task share the universal prompt. However, a fine-grained prompt design can enhance the performance of few-shot learning task by leveraging more diverse information hidden in the set of examples. In light of this motivation, this paper introduce an example-specific prompt learning method to embody fine-grained self-adapting prompts for few-shot learning with pre-trained models. Specifically, we introduce the concept of the “weak consistency assumption”, to trade-off the task-specific consistent and example-specific diversity. Based on this assumption, a novel method called Self-adapting Continuous Prompt Learning (SP-learning) to learn example-specific prompts is proposed. It employs a cross-attention prompt generator that considers the characteristics of input samples and utilizes a diversity calibration technique to adjust the prompt generator accordingly. By personalizing prompts for each example, SP-learning aims to improve few-shot learning performance. We perform a systematic evaluation on 10 public benchmark tasks and our method outperforms 8 of those tasks. Our research sheds light on the importance of personalized prompts and opens up new possibilities for improving few-shot learning tasks.

Few-shot image classification via hybrid representation

Few-shot image classification aims to learn an embedding model on the base datasets and design a base learner to recognize novel categories. The few-shot image classification framework is a two-phase process. First, the pre-train phase utilizes the base data to train a CNN-based feature extractor. Next, in the meta-test phase, the frozen feature extractor is applied to novel data with categories different from the base data. A base learner is then designed for recognition. Several simple base learners, including nearest neighbor, support vector machine, and logistic regression classifiers, have been recently introduced for few-shot learning tasks. However, these base learners are separately designed to consider specific representations (e.g., the class center) or shared representations (e.g., the boundaries). This paper mainly focuses on exploring the representation-residual base learners, which aim to represent a query sample with the support set and predict the query sample’s label based on the minimal residual error. We first introduce two representation-residual base learners: a specific representation base learner and a shared representation base learner. Then, we propose a novel hybrid representation base learner that combines both base learners to generate competitive representation. Additionally, we extend our approach by incorporating a self-training framework to utilize the query data fully. We evaluate our proposed method on several benchmark few-shot image classification datasets, such as miniImageNet, tieredImageNet, CIFAR-FS, FC100, and CUB datasets. The experimental results indicate that our proposed approach shows a significant performance improvement.

Few-shot defect classification via feature aggregation based on graph neural network

The effectiveness of deep learning models is greatly dependent on the availability of a vast amount of labeled data. However, in the realm of surface defect classification, acquiring and annotating defect samples proves to be quite challenging. Consequently, accurately predicting defect types with only a limited number of labeled samples has emerged as a prominent research focus in recent years. Few-shot learning, which leverages a restricted sample set in the support set, can effectively predict the categories of unlabeled samples in the query set. This approach is particularly well-suited for defect classification scenarios. In this article, we propose a transductive few-shot surface defect classification method, which using both the instance-level relations and distribution-level relations in each few-shot learning task. Furthermore, we calculate class center features in transductive manner and incorporate them into the feature aggregation operation to rectify the positioning of edge samples in the mapping space. This adjustment aims to minimize the distance between samples of the same category, thereby mitigating the influence of unlabeled samples at category boundary on classification accuracy. Experimental results on the public dataset show the outstanding performance of our proposed approach compared to the state-of-the-art methods in the few-shot learning settings. Our code is available at https://github.com/Harry10459/CIDnet.

Dual-Level Curriculum Meta-Learning for Noisy Few-Shot Learning Tasks

Few-shot learning (FSL) is essential in many practical applications. However, the limited training examples make the models more vulnerable to label noise, which can lead to poor generalization capability. To address this critical challenge, we propose a curriculum meta-learning model that employs a novel dual-level class-example sampling strategy to create a robust curriculum for adaptive task distribution formulation and robust model training. The dual-level framework proposes a heuristic class sampling criterion that measures pairwise class boundary complexity to form a class curriculum; it uses effective example sampling through an under-trained proxy model to form an example curriculum. By utilizing both class-level and example-level information, our approach is more robust to handle limited training data and noisy labels that commonly occur in few-shot learning tasks. The model has efficient convergence behavior, which is verified through rigorous convergence analysis. Additionally, we establish a novel error bound through a hierarchical PAC-Bayesian analysis for curriculum meta-learning under noise. We conduct extensive experiments that demonstrate the effectiveness of our framework in outperforming existing noisy few-shot learning methods under various few-shot classification benchmarks. Our code is available at https://github.com/ritmininglab/DCML.

MetaDiff: Meta-Learning with Conditional Diffusion for Few-Shot Learning

Equipping a deep model the ability of few-shot learning (FSL) is a core challenge for artificial intelligence. Gradient-based meta-learning effectively addresses the challenge by learning how to learn novel tasks. Its key idea is learning a deep model in a bi-level optimization manner, where the outer-loop process learns a shared gradient descent algorithm (called meta-optimizer), while the inner-loop process leverages it to optimize a task-specific base learner with few examples. Although these methods have shown superior performance on FSL, the outer-loop process requires calculating second-order derivatives along the inner-loop path, which imposes considerable memory burdens and the risk of vanishing gradients. This degrades meta-learning performance. Inspired by recent diffusion models, we find that the inner-loop gradient descent process can be viewed as a reverse process (i.e., denoising) of diffusion where the target of denoising is the weight of base learner but origin data. Based on this fact, we propose to model the gradient descent algorithm as a diffusion model and then present a novel conditional diffusion-based meta-learning, called MetaDiff, that effectively models the optimization process of base learner weights from Gaussian initialization to target weights in a denoising manner. Thanks to the training efficiency of diffusion models, our MetaDiff does not need to differentiate through the inner-loop path such that the memory burdens and the risk of vanishing gradients can be effectively alleviated for improving FSL. Experimental results show that our MetaDiff outperforms state-of-the-art gradient-based meta-learning family on FSL tasks.

Does Few-Shot Learning Suffer from Backdoor Attacks?

The field of few-shot learning (FSL) has shown promising results in scenarios where training data is limited, but its vulnerability to backdoor attacks remains largely unexplored. We first explore this topic by first evaluating the performance of the existing backdoor attack methods on few-shot learning scenarios. Unlike in standard supervised learning, existing backdoor attack methods failed to perform an effective attack in FSL due to two main issues. Firstly, the model tends to overfit to either benign features or trigger features, causing a tough trade-off between attack success rate and benign accuracy. Secondly, due to the small number of training samples, the dirty label or visible trigger in the support set can be easily detected by victims, which reduces the stealthiness of attacks. It seemed that FSL could survive from backdoor attacks. However, in this paper, we propose the Few-shot Learning Backdoor Attack (FLBA) to show that FSL can still be vulnerable to backdoor attacks. Specifically, we first generate a trigger to maximize the gap between poisoned and benign features. It enables the model to learn both benign and trigger features, which solves the problem of overfitting. To make it more stealthy, we hide the trigger by optimizing two types of imperceptible perturbation, namely attractive and repulsive perturbation, instead of attaching the trigger directly. Once we obtain the perturbations, we can poison all samples in the benign support set into a hidden poisoned support set and fine-tune the model on it. Our method demonstrates a high Attack Success Rate (ASR) in FSL tasks with different few-shot learning paradigms while preserving clean accuracy and maintaining stealthiness. This study reveals that few-shot learning still suffers from backdoor attacks, and its security should be given attention.

Few-Shot Learning via Repurposing Ensemble of Black-Box Models

This paper investigates the problem of exploiting existing solution models of previous tasks to address a related target task with limited training data. Existing approaches addressing this problem often require access to the internal parameterization of the existing solution models and possibly their training data, which is not possible in many practical settings. To relax this requirement, We approach this problem from a new perspective of black-box re-purposing, which augments the target inputs and leverages their corresponding outputs generated by existing black-box APIs into a feature ensemble. We hypothesize that such feature ensemble can be learned to incorporate and encode relevant black-box knowledge into the feature representation of target data, which will compensate for their scarcity. This hypothesis is confirmed via the reported successes of our proposed black-box ensemble in solving multiple few-shot learning tasks derived from various benchmark datasets. All reported results show consistently that the set of heterogeneous black-box solutions of previous tasks can indeed be reused and combined effectively to solve a reasonably related target task without requiring access to a large training dataset. This is the first step towards enabling new possibilities to further supplement existing techniques in transfer or meta learning with black-box knowledge.

Tagging Items with Emerging Tags: A Neural Topic Model Based Few-Shot Learning Approach

The tagging system has become a primary tool to organize information resources on the Internet, which benefits both users and the platforms. To build a successful tagging system, automatic tagging methods are desired. With the development of society, new tags keep emerging. The problem of tagging items with emerging tags is an open challenge for an automatic tagging system, and it has not been well studied in the literature. We define this problem as a tag-centered cold-start problem in this study and propose a novel neural topic model based few-shot learning method named NTFSL to solve the problem. In our proposed method, we innovatively fuse the topic modeling task with the few-shot learning task, endowing the model with the capability to infer effective topics to solve the tag-centered cold-start problem with the property of interpretability. Meanwhile, we propose a novel neural topic model for the topic modeling task to improve the quality of inferred topics, which helps enhance the tagging performance. Furthermore, we develop a novel inference method based on the variational auto-encoding framework for model inference. We conducted extensive experiments on two real-world datasets, and the results demonstrate the superior performance of our proposed model compared with state-of-the-art machine learning methods. Case studies also show the interpretability of the model.

Emb-trattunet: a novel edge loss function and transformer-CNN architecture for multi-classes pneumonia infection segmentation in low annotation regimes

One of the primary challenges in applying deep learning approaches to medical imaging is the limited availability of data due to various factors. These factors include concerns about data privacy and the requirement for expert radiologists to perform the time-consuming and labor-intensive task of labeling data, particularly for tasks such as segmentation. Consequently, there is a critical need to develop novel approaches for few-shot learning tasks in this domain. In this work, we propose a Novel CNN-Transformer Fusion scheme to segment Multi-classes pneumonia infection from limited CT-scans data. In total, there are three main contributions: (i) CNN-Transformer encoders fusion, which allows to extract and fuse richer features in the encoding phase, which contains: local, global and long-range dependencies features, (ii) Multi-Branches Skip Connection (MBSC) is proposed to extract and fuse richer features from the encoder features then integrate them into the decoder layers, where MBSC blocks extract higher-level features related to the finer details of different infection types, and (iii) a Multi-classes Boundary Aware Cross-Entropy (MBA-CE) Loss function is proposed to deal with fuzzy boundaries, enhance the separability between classes and give more attention to the minority classes. The performance of the proposed approach is evaluated using two evaluation scenarios and compared with different baseline and state-of-the-art segmentation architectures for Multi-classes Covid-19 segmentation. The obtained results show that our approach outperforms the comparison methods in both Ground-Glass Opacity (GGO) and Consolidation segmentation. On the other hand, our approach shows consistent performance when the training data is reduced to half, which proves the efficiency of our approach in few-shot learning. In contrast, the performance of the comparison methods drops in this scenario. Moreover, our approach is able to deal with imbalanced data classes. These advantages prove the effectiveness and efficiency of the proposed EMB-TrAttUnet approach in a pandemic scenario where time is critical to save patient lives.

Edge-labeling based modified gated graph network for few-shot learning

Accurate determination of similarity between samples is fundamental and critical for graph network based few-shot learning tasks. Previous approaches typically employ convolutional neural networks to obtain relations between nodes. However, these networks are not adept at handling node features in vector form. To overcome this limitation, we proposed a modified gated graph network (MGGN) that uniquely integrates graph networks and modified gated recurrent units (M-GRU) for few-shot classification. The introduced M-GRU mitigates the loss of label information from the initial graph and reduces computational complexity. The MGGN contains two modules that alternately update node and edge features. The node update module leverages a gating mechanism to integrate edge features into node update weights, fostering a learnable node aggregation process. The edge update component perceives the trend in edge feature changes and establishes long-term dependencies. Experimental results on two benchmark datasets demonstrate that our MGGN achieves comparable performance to state-of-the-art methods. The code is available at https://github.com/zpx16900/MGGN.

MMT: Cross Domain Few-Shot Learning via Meta-Memory Transfer.

Few-shot learning aims to recognize novel categories solely relying on a few labeled samples, with existing few-shot methods primarily focusing on the categories sampled from the same distribution. Nevertheless, this assumption cannot always be ensured, and the actual domain shift problem significantly reduces the performance of few-shot learning. To remedy this problem, we investigate an interesting and challenging cross-domain few-shot learning task, where the training and testing tasks employ different domains. Specifically, we propose a Meta-Memory scheme to bridge the domain gap between source and target domains, leveraging style-memory and content-memory components. The former stores intra-domain style information from source domain instances and provides a richer feature distribution. The latter stores semantic information through exploration of knowledge of different categories. Under the contrastive learning strategy, our model effectively alleviates the cross-domain problem in few-shot learning. Extensive experiments demonstrate that our proposed method achieves state-of-the-art performance on cross-domain few-shot semantic segmentation tasks on the COCO-20 i, PASCAL-5 i, FSS-1000, and SUIM datasets and positively affects few-shot classification tasks on Meta-Dataset.

Dynamic WiFi indoor positioning based on the multi-scale metric learning

At present, most of the fingerprint-based positioning methods require the target to stay static in the process of data collection, which are not applicable to the target positioning in a moving state. Common dynamic localization and tracking algorithms such as Kalman Filter (KF) and Particle Filter (PF) need to estimate the position at the current time according to the estimated position at the previous time. If there is a positioning error at a certain time, the following positioning performance will be affected by the cumulative errors. To solve these problems, this paper proposes a dynamic indoor positioning method based on the multi-scale metric learning of the channel state information (CSI), named as CSI-MML. This method can realize dynamic positioning using CSI signals without carrying extra equipment. The proposed model of CSI-MML is trained and tested by constructing few-shot learning tasks, which is mainly composed of two parts: feature extraction and similarity metric. In order to fully extract the effective features of the samples, the attention mechanism is added to the feature extraction module of the network to enhance the model’s ability of extracting significant features. The similarity metric module measures the global similarity and local similarity between samples and fuses the two similarities by a similarity fusion layer. Through the multi-scale metric, we can improve the metric ability of CSI-MML, so as to improve the positioning accuracy of dynamic positioning. The experimental results show that compared with the commonly used dynamic location and tracking algorithms such as KF and PF, the positioning error of the proposed method will not accumulate and the average positioning error is smaller.