With the introduction of the PSD2 regulation in the EU which established the Open Banking framework, a new window of opportunities has opened for banks and fintechs to explore and enrich Bank transaction descriptions with the aim of building a better understanding of customer behavior, while using this understanding to prevent fraud, reduce risks and offer more competitive and tailored services. And although the usage of natural language processing models and techniques has seen an incredible progress in various applications and domains over the past few years, custom applications based on domain-specific text corpus remain unaddressed especially in the banking sector. In this paper, we introduce a language-based Open Banking transaction classification system with a focus on the french market and french language text. The system encompasses data collection, labeling, preprocessing, modeling, and evaluation stages. Unlike previous studies that focus on general classification approaches, this system is specifically tailored to address the challenges posed by training a language model with a specialized text corpus (Banking data in the French context). By incorporating language-specific techniques and domain knowledge, the proposed system demonstrates enhanced performance and efficiency compared to generic approaches.
This work invokes the notion of $f$-divergence to introduce a novel upper bound on the Bayes error rate of a general classification task. We show that the proposed bound can be computed by sampling from the output of a parameterized model. Using this practical interpretation, we introduce the Bayes optimal learning threshold (BOLT) loss whose minimization enforces a classification model to achieve the Bayes error rate. We validate the proposed loss for image and text classification tasks, considering MNIST, Fashion-MNIST, CIFAR-10, and IMDb datasets. Numerical experiments demonstrate that models trained with BOLT achieve performance on par with or exceeding that of cross-entropy, particularly on challenging datasets. This highlights the potential of BOLT in improving generalization.
With the rapid advancement of Multimodal Large Language Models (MLLMs), a variety of benchmarks have been introduced to evaluate their capabilities. While most evaluations have focused on complex tasks such as scientific comprehension and visual reasoning, little attention has been given to assessing their fundamental image classification abilities. In this paper, we address this gap by thoroughly revisiting the MLLMs with an in-depth analysis of image classification. Specifically, building on established datasets, we examine a broad spectrum of scenarios, from general classification tasks (e.g., ImageNet, ObjectNet) to more fine-grained categories such as bird and food classification. Our findings reveal that the most recent MLLMs can match or even outperform CLIP-style vision-language models on several datasets, challenging the previous assumption that MLLMs are bad at image classification \cite{VLMClassifier}. To understand the factors driving this improvement, we conduct an in-depth analysis of the network architecture, data selection, and training recipe used in public MLLMs. Our results attribute this success to advancements in language models and the diversity of training data sources. Based on these observations, we further analyze and attribute the potential reasons to conceptual knowledge transfer and enhanced exposure of target concepts, respectively. We hope our findings will offer valuable insights for future research on MLLMs and their evaluation in image classification tasks.
Objectives: Compare qualitative coding of instruction tuned large language models (IT-LLMs) against human coders in classifying the presence or absence of vulnerability in routinely collected unstructured text that describes police-public interactions. Evaluate potential bias in IT-LLM codings. Methods: Analyzing publicly available text narratives of police-public interactions recorded by Boston Police Department, we provide humans and IT-LLMs with qualitative labelling codebooks and compare labels generated by both, seeking to identify situations associated with (i) mental ill health; (ii) substance misuse; (iii) alcohol dependence; and (iv) homelessness. We explore multiple prompting strategies and model sizes, and the variability of labels generated by repeated prompts. Additionally, to explore model bias, we utilize counterfactual methods to assess the impact of two protected characteristics - race and gender - on IT-LLM classification. Results: Results demonstrate that IT-LLMs can effectively support human qualitative coding of police incident narratives. While there is some disagreement between LLM and human generated labels, IT-LLMs are highly effective at screening narratives where no vulnerabilities are present, potentially vastly reducing the requirement for human coding. Counterfactual analyses demonstrate that manipulations to both gender and race of individuals described in narratives have very limited effects on IT-LLM classifications beyond those expected by chance. Conclusions: IT-LLMs offer effective means to augment human qualitative coding in a way that requires much lower levels of resource to analyze large unstructured datasets. Moreover, they encourage specificity in qualitative coding, promote transparency, and provide the opportunity for more standardized, replicable approaches to analyzing large free-text police data sources.
Recently, a number of image-mixing-based augmentation techniques have been introduced to improve the generalization of deep neural networks. In these techniques, two or more randomly selected natural images are mixed together to generate an augmented image. Such methods may not only omit important portions of the input images but also introduce label ambiguities by mixing images across labels resulting in misleading supervisory signals. To address these limitations, we propose DiffuseMix, a novel data augmentation technique that leverages a diffusion model to reshape training images, supervised by our bespoke conditional prompts. First, concatenation of a partial natural image and its generated counterpart is obtained which helps in avoiding the generation of unrealistic images or label ambiguities. Then, to enhance resilience against adversarial attacks and improves safety measures, a randomly selected structural pattern from a set of fractal images is blended into the concatenated image to form the final augmented image for training. Our empirical results on seven different datasets reveal that DiffuseMix achieves superior performance compared to existing state-of the-art methods on tasks including general classification,fine-grained classification, fine-tuning, data scarcity, and adversarial robustness. Augmented datasets and codes are available here: https://diffusemix.github.io/
This work continues to investigate the link between differentially private (DP) and online learning. Alon, Livni, Malliaris, and Moran (2019) showed that for binary concept classes, DP learnability of a given class implies that it has a finite Littlestone dimension (equivalently, that it is online learnable). Their proof relies on a model-theoretic result by Hodges (1997), which demonstrates that any binary concept class with a large Littlestone dimension contains a large subclass of thresholds. In a follow-up work, Jung, Kim, and Tewari (2020) extended this proof to multiclass PAC learning with a bounded number of labels. Unfortunately, Hodges's result does not apply in other natural settings such as multiclass PAC learning with an unbounded label space, and PAC learning of partial concept classes. This naturally raises the question of whether DP learnability continues to imply online learnability in more general scenarios: indeed, Alon, Hanneke, Holzman, and Moran (2021) explicitly leave it as an open question in the context of partial concept classes, and the same question is open in the general multiclass setting. In this work, we give a positive answer to these questions showing that for general classification tasks, DP learnability implies online learnability. Our proof reasons directly about Littlestone trees, without relying on thresholds. We achieve this by establishing several Ramsey-type theorems for trees, which might be of independent interest.
Human action video recognition has recently attracted more attention in applications such as video security and sports posture correction. Popular solutions, including graph convolutional networks (GCNs) that model the human skeleton as a spatiotemporal graph, have proven very effective. GCNs-based methods with stacked blocks usually utilize top-layer semantics for classification/annotation purposes. Although the global features learned through the procedure are suitable for the general classification, they have difficulty capturing fine-grained action change across adjacent frames -- decisive factors in sports actions. In this paper, we propose a novel ``Cross-block Fine-grained Semantic Cascade (CFSC)'' module to overcome this challenge. In summary, the proposed CFSC progressively integrates shallow visual knowledge into high-level blocks to allow networks to focus on action details. In particular, the CFSC module utilizes the GCN feature maps produced at different levels, as well as aggregated features from proceeding levels to consolidate fine-grained features. In addition, a dedicated temporal convolution is applied at each level to learn short-term temporal features, which will be carried over from shallow to deep layers to maximize the leverage of low-level details. This cross-block feature aggregation methodology, capable of mitigating the loss of fine-grained information, has resulted in improved performance. Last, FD-7, a new action recognition dataset for fencing sports, was collected and will be made publicly available. Experimental results and empirical analysis on public benchmarks (FSD-10) and self-collected (FD-7) demonstrate the advantage of our CFSC module on learning discriminative patterns for action classification over others.
We consider gradient descent (GD) with a constant stepsize applied to logistic regression with linearly separable data, where the constant stepsize $\eta$ is so large that the loss initially oscillates. We show that GD exits this initial oscillatory phase rapidly -- in $\mathcal{O}(\eta)$ steps -- and subsequently achieves an $\tilde{\mathcal{O}}(1 / (\eta t) )$ convergence rate after $t$ additional steps. Our results imply that, given a budget of $T$ steps, GD can achieve an accelerated loss of $\tilde{\mathcal{O}}(1/T^2)$ with an aggressive stepsize $\eta:= \Theta( T)$, without any use of momentum or variable stepsize schedulers. Our proof technique is versatile and also handles general classification loss functions (where exponential tails are needed for the $\tilde{\mathcal{O}}(1/T^2)$ acceleration), nonlinear predictors in the neural tangent kernel regime, and online stochastic gradient descent (SGD) with a large stepsize, under suitable separability conditions.
Facial Beauty Prediction (FBP) is a computer vision task of quantifying the beauty of a face. Several solutions to this problem have benefitted immensely from the recent developments in deep learning. However, the majority of current methods train machine learning models to purely predict mean beauty scores, treating FBP solely as a regression task. In addition, deep learning based FBP approaches so far use transfer learning from models trained on general classification tasks such as ImageNet. We propose fine-tuning an ensemble of convolutional neural network (CNN) models originally trained on face verification tasks using a variety of loss functions such as Earth Mover's Distance (EMD) based loss. With this approach, our method can predict the entire beauty score distribution rather than just the mean, and the predicted mean scores have a higher Pearson Correlation (PC) compared to the ground truth scores. This method achieves state of the art results on the MEBeauty dataset in terms of mean absolute error, root mean squared error and PC between the predicted and the ground truth mean scores.
Unlike the more commonly analyzed ECG or PPG data for activity classification, heart rate time series data is less detailed, often noisier and can contain missing data points. Using the BigIdeasLab_STEP dataset, which includes heart rate time series annotated with specific tasks performed by individuals, we sought to determine if general classification was achievable. Our analyses showed that the accuracy is sensitive to the choice of window/stride size. Moreover, we found variable classification performances between subjects due to differences in the physical structure of their hearts. Various techniques were used to minimize this variability. First of all, normalization proved to be a crucial step and significantly improved the performance. Secondly, grouping subjects and performing classification inside a group helped to improve performance and decrease inter-subject variability. Finally, we show that including handcrafted features as input to a deep learning (DL) network improves the classification performance further. Together, these findings indicate that heart rate time series can be utilized for classification tasks like predicting activity. However, normalization or grouping techniques need to be chosen carefully to minimize the issue of subject variability.
Joint intent detection and slot filling, which is also termed as joint NLU (Natural Language Understanding) is invaluable for smart voice assistants. Recent advancements in this area have been heavily focusing on improving accuracy using various techniques. Explainability is undoubtedly an important aspect for deep learning-based models including joint NLU models. Without explainability, their decisions are opaque to the outside world and hence, have tendency to lack user trust. Therefore to bridge this gap, we transform the full joint NLU model to be `inherently' explainable at granular levels without compromising on accuracy. Further, as we enable the full joint NLU model explainable, we show that our extension can be successfully used in other general classification tasks. We demonstrate this using sentiment analysis and named entity recognition.
The recent surge of interest surrounding Multimodal Neural Networks (MM-NN) is attributed to their ability to effectively process and integrate multiscale information from diverse data sources. MM-NNs extract and fuse features from multiple modalities using adequate unimodal backbones and specific fusion networks. Although this helps strengthen the multimodal information representation, designing such networks is labor-intensive. It requires tuning the architectural parameters of the unimodal backbones, choosing the fusing point, and selecting the operations for fusion. Furthermore, multimodality AI is emerging as a cutting-edge option in Internet of Things (IoT) systems where inference latency and energy consumption are critical metrics in addition to accuracy. In this paper, we propose Harmonic-NAS, a framework for the joint optimization of unimodal backbones and multimodal fusion networks with hardware awareness on resource-constrained devices. Harmonic-NAS involves a two-tier optimization approach for the unimodal backbone architectures and fusion strategy and operators. By incorporating the hardware dimension into the optimization, evaluation results on various devices and multimodal datasets have demonstrated the superiority of Harmonic-NAS over state-of-the-art approaches achieving up to 10.9% accuracy improvement, 1.91x latency reduction, and 2.14x energy efficiency gain.
Preference-based reinforcement learning (PbRL) promises to learn a complex reward function with binary human preference. However, such human-in-the-loop formulation requires considerable human effort to assign preference labels to segment pairs, hindering its large-scale applications. Recent approache has tried to reuse unlabeled segments, which implicitly elucidates the distribution of segments and thereby alleviates the human effort. And consistency regularization is further considered to improve the performance of semi-supervised learning. However, we notice that, unlike general classification tasks, in PbRL there exits a unique phenomenon that we defined as similarity trap in this paper. Intuitively, human can have diametrically opposite preferredness for similar segment pairs, but such similarity may trap consistency regularization fail in PbRL. Due to the existence of similarity trap, such consistency regularization improperly enhances the consistency possiblity of the model's predictions between segment pairs, and thus reduces the confidence in reward learning, since the augmented distribution does not match with the original one in PbRL. To overcome such issue, we present a self-training method along with our proposed peer regularization, which penalizes the reward model memorizing uninformative labels and acquires confident predictions. Empirically, we demonstrate that our approach is capable of learning well a variety of locomotion and robotic manipulation behaviors using different semi-supervised alternatives and peer regularization.
Nowadays, there is growing interest in applying Artificial Intelligence (AI) on board Earth Observation (EO) satellites for time-critical applications, such as natural disaster response. However, the unavailability of raw satellite data currently hinders research on lightweight pre-processing techniques and limits the exploration of end-to-end pipelines, which could offer more efficient and accurate extraction of insights directly from the source data. To fill this gap, this work presents a novel methodology to automate the creation of datasets for the detection of target events (e.g., warm thermal hotspots) or objects (e.g., vessels) from Sentinel-2 raw data and other multispectral EO pushbroom raw imagery. The presented approach first processes the raw data by applying a pipeline consisting of spatial band registration and georeferencing of the raw data pixels. Then, it detects the target events by leveraging event-specific state-of-the-art algorithms on the Level-1C products, which are mosaicked and cropped on the georeferenced correspondent raw granule area. The detected events are finally re-projected back onto the corresponding raw images. We apply the proposed methodology to realize THRawS (Thermal Hotspots in Raw Sentinel-2 data), the first dataset of Sentinel-2 raw data containing warm thermal hotspots. THRawS includes 1090 samples containing wildfires, volcanic eruptions, and 33,335 event-free acquisitions to enable thermal hotspot detection and general classification applications. This dataset and associated toolkits provide the community with both an immediately useful resource as well as a framework and methodology acting as a template for future additions. With this work, we hope to pave the way for research on energy-efficient pre-processing algorithms and AI-based end-to-end processing systems on board EO satellites.
Text clustering, as one of the most fundamental challenges in unsupervised learning, aims at grouping semantically similar text segments without relying on human annotations. With the rapid development of deep learning, deep clustering has achieved significant advantages over traditional clustering methods. Despite the effectiveness, most existing deep text clustering methods rely heavily on representations pre-trained in general domains, which may not be the most suitable solution for clustering in specific target domains. To address this issue, we propose CEIL, a novel Classification-Enhanced Iterative Learning framework for short text clustering, which aims at generally promoting the clustering performance by introducing a classification objective to iteratively improve feature representations. In each iteration, we first adopt a language model to retrieve the initial text representations, from which the clustering results are collected using our proposed Category Disentangled Contrastive Clustering (CDCC) algorithm. After strict data filtering and aggregation processes, samples with clean category labels are retrieved, which serve as supervision information to update the language model with the classification objective via a prompt learning approach. Finally, the updated language model with improved representation ability is used to enhance clustering in the next iteration. Extensive experiments demonstrate that the CEIL framework significantly improves the clustering performance over iterations, and is generally effective on various clustering algorithms. Moreover, by incorporating CEIL on CDCC, we achieve the state-of-the-art clustering performance on a wide range of short text clustering benchmarks outperforming other strong baseline methods.
Since the introduction of deep learning, a wide scope of representation properties, such as decorrelation, whitening, disentanglement, rank, isotropy, and mutual information, have been studied to improve the quality of representation. However, manipulating such properties can be challenging in terms of implementational effectiveness and general applicability. To address these limitations, we propose to regularize von Neumann entropy~(VNE) of representation. First, we demonstrate that the mathematical formulation of VNE is superior in effectively manipulating the eigenvalues of the representation autocorrelation matrix. Then, we demonstrate that it is widely applicable in improving state-of-the-art algorithms or popular benchmark algorithms by investigating domain-generalization, meta-learning, self-supervised learning, and generative models. In addition, we formally establish theoretical connections with rank, disentanglement, and isotropy of representation. Finally, we provide discussions on the dimension control of VNE and the relationship with Shannon entropy. Code is available at: https://github.com/jaeill/CVPR23-VNE.
This paper introduces a general model called CIPNN - Continuous Indeterminate Probability Neural Network, and this model is based on IPNN, which is used for discrete latent random variables. Currently, posterior of continuous latent variables is regarded as intractable, with the new theory proposed by IPNN this problem can be solved. Our contributions are Four-fold. First, we derive the analytical solution of the posterior calculation of continuous latent random variables and propose a general classification model (CIPNN). Second, we propose a general auto-encoder called CIPAE - Continuous Indeterminate Probability Auto-Encoder, the decoder part is not a neural network and uses a fully probabilistic inference model for the first time. Third, we propose a new method to visualize the latent random variables, we use one of N dimensional latent variables as a decoder to reconstruct the input image, which can work even for classification tasks, in this way, we can see what each latent variable has learned. Fourth, IPNN has shown great classification capability, CIPNN has pushed this classification capability to infinity. Theoretical advantages are reflected in experimental results.
In recent years, the rapid rise of video applications has led to an explosion of Internet video traffic, thereby posing severe challenges to network management. Therefore, effectively identifying and managing video traffic has become an urgent problem to be solved. However, the existing video traffic feature extraction methods mainly target at the traditional packet and flow level features, and the video traffic identification accuracy is low. Additionally, the issue of high data dimension often exists in video traffic identification, requiring an effective approach to select the most relevant features to complete the identification task. Although numerous studies have used feature selection to achieve improved identification performance, no feature selection research has focused on measuring feature distributions that do not overlap or have a small overlap. First, this study proposes to extract video-related features to construct a large-scale feature set to identify video traffic. Second, to reduce the cost of video traffic identification and select an effective feature subset, the current research proposes an adaptive distribution distance-based feature selection (ADDFS) method, which uses Wasserstein distance to measure the distance between feature distributions. To test the effectiveness of the proposal, we collected a set of video traffic from different platforms in a campus network environment and conducted a set of experiments using these data sets. Experimental results suggest that the proposed method can achieve high identification performance for video scene traffic and cloud game video traffic identification. Lastly, a comparison of ADDFS with other feature selection methods shows that ADDFS is a practical feature selection technique not only for video traffic identification, but also for general classification tasks.
Science and Engineering applications are typically associated with expensive optimization problems to identify optimal design solutions and states of the system of interest. Bayesian optimization and active learning compute surrogate models through efficient adaptive sampling schemes to assist and accelerate this search task toward a given optimization goal. Both those methodologies are driven by specific infill/learning criteria which quantify the utility with respect to the set goal of evaluating the objective function for unknown combinations of optimization variables. While the two fields have seen an exponential growth in popularity in the past decades, their dualism and synergy have received relatively little attention to date. This paper discusses and formalizes the synergy between Bayesian optimization and active learning as symbiotic adaptive sampling methodologies driven by common principles. In particular, we demonstrate this unified perspective through the formalization of the analogy between the Bayesian infill criteria and active learning criteria as driving principles of both the goal-driven procedures. To support our original perspective, we propose a general classification of adaptive sampling techniques to highlight similarities and differences between the vast families of adaptive sampling, active learning, and Bayesian optimization. Accordingly, the synergy is demonstrated mapping the Bayesian infill criteria with the active learning criteria, and is formalized for searches informed by both a single information source and multiple levels of fidelity. In addition, we provide guidelines to apply those learning criteria investigating the performance of different Bayesian schemes for a variety of benchmark problems to highlight benefits and limitations over mathematical properties that characterize real-world applications.
The electrocardiogram (ECG) is one of the most common primary tests to evaluate the health of the heart. Reliable automatic interpretation of ECG records is crucial to the goal of improving public health. It can enable a safe inexpensive monitoring. This work presents a new methodology for morphological identification of heartbeats, which is placed outside the usual machine learning framework. The proposal considers the sparsity of the representation of a heartbeat as a parameter for morphological identification. The approach involves greedy algorithms for selecting elements from redundant dictionaries, which should be previously learnt from examples of the classes to be identified. Using different metrics of sparsity, the dictionary rendering the smallest sparsity value, for the equivalent approximation quality of a new heartbeat, classifies the morphology of that beat. This study focuses on a procedure of learning the dictionaries for representing heartbeats and compares several metrics of sparsity for morphological identification on the basis of those metrics. The suitability of the method is illustrated by binary differentiation of Normal and Ventricular heartbeats in the MIT-BIH Arrhythmia data set. In general classification 99.7% of the Normal beats and 97.6% of the Ventricular beats in the testing sets are correctly identified. In interpatient assessment 91.8% of the Normal beats and 91.0% of Ventricular beats are correctly identified. Even more important than these scores is the fact that they are produced on the bases of a single parameter. The numerical tests, designed to emphasise the interpretability and reliability of the approach, demonstrate the potential of the method to contribute towards the development of a well grounded expert system for classification of heartbeats in ECG records.