Accurate automatic screening of minors in unconstrained images demands models that are robust to distribution shift and resilient to the children under-representation in publicly available data. To overcome these issues, we propose a multi-task architecture with dedicated under/over-age discrimination tasks based on a frozen FaRL vision-language backbone joined with a compact two-layer MLP that shares features across one age-regression head and four binary under-age heads for age thresholds of 12, 15, 18, and 21 years, focusing on the legally critical age range. To address the severe class imbalance, we introduce an $\alpha$-reweighted focal-style loss and age-balanced mini-batch sampling, which equalizes twelve age bins during stochastic optimization. Further improvement is achieved with an age gap that removes edge cases from the loss. Moreover, we set a rigorous evaluation by proposing the Overall Under-Age Benchmark, with 303k cleaned training images and 110k test images, defining both the "ASORES-39k" restricted overall test, which removes the noisiest domains, and the age estimation wild shifts test "ASWIFT-20k" of 20k-images, stressing extreme pose ($>$45{\deg}), expression, and low image quality to emulate real-world shifts. Trained on the cleaned overall set with resampling and age gap, our multiage model "F" lowers the root-mean-square-error on the ASORES-39k restricted test from 5.733 (age-only baseline) to 5.656 years and lifts under-18 detection from F2 score of 0.801 to 0.857 at 1% false-adult rate. Under the domain shift to the wild data of ASWIFT-20k, the same configuration nearly sustains 0.99 recall while boosting F2 from 0.742 to 0.833 with respect to the age-only baseline, demonstrating strong generalization under distribution shift. For the under-12 and under-15 tasks, the respective boosts in F2 are from 0.666 to 0.955 and from 0.689 to 0.916, respectively.
The advent of Open Radio Access Networks (O-RAN) has transformed the telecommunications industry by promoting interoperability, vendor diversity, and rapid innovation. However, its disaggregated architecture introduces complex testing challenges, particularly in validating multi-vendor components against O-RAN ALLIANCE and 3GPP specifications. Existing frameworks, such as those provided by Open Testing and Integration Centres (OTICs), rely heavily on manual processes, are fragmented and prone to human error, leading to inconsistency and scalability issues. To address these limitations, we present AI5GTest -- an AI-powered, specification-aware testing framework designed to automate the validation of O-RAN components. AI5GTest leverages a cooperative Large Language Models (LLM) framework consisting of Gen-LLM, Val-LLM, and Debug-LLM. Gen-LLM automatically generates expected procedural flows for test cases based on 3GPP and O-RAN specifications, while Val-LLM cross-references signaling messages against these flows to validate compliance and detect deviations. If anomalies arise, Debug-LLM performs root cause analysis, providing insight to the failure cause. To enhance transparency and trustworthiness, AI5GTest incorporates a human-in-the-loop mechanism, where the Gen-LLM presents top-k relevant official specifications to the tester for approval before proceeding with validation. Evaluated using a range of test cases obtained from O-RAN TIFG and WG5-IOT test specifications, AI5GTest demonstrates a significant reduction in overall test execution time compared to traditional manual methods, while maintaining high validation accuracy.
Deep models, such as convolutional neural networks (CNNs) and vision transformer (ViT), demonstrate remarkable performance in image classification. However, those deep models require large data to fine-tune, which is impractical in the medical domain due to the data privacy issue. Furthermore, despite the feasible performance of contrastive language image pre-training (CLIP) in the natural domain, the potential of CLIP has not been fully investigated in the medical field. To face these challenges, we considered three scenarios: 1) we introduce a novel CLIP variant using four CNNs and eight ViTs as image encoders for the classification of brain cancer and skin cancer, 2) we combine 12 deep models with two federated learning techniques to protect data privacy, and 3) we involve traditional machine learning (ML) methods to improve the generalization ability of those deep models in unseen domain data. The experimental results indicate that maxvit shows the highest averaged (AVG) test metrics (AVG = 87.03\%) in HAM10000 dataset with multimodal learning, while convnext\_l demonstrates remarkable test with an F1-score of 83.98\% compared to swin\_b with 81.33\% in FL model. Furthermore, the use of support vector machine (SVM) can improve the overall test metrics with AVG of $\sim 2\%$ for swin transformer series in ISIC2018. Our codes are available at https://github.com/AIPMLab/SkinCancerSimulation.
Spontaneous speech emotion data usually contain perceptual grades where graders assign emotion score after listening to the speech files. Such perceptual grades introduce uncertainty in labels due to grader opinion variation. Grader variation is addressed by using consensus grades as groundtruth, where the emotion with the highest vote is selected. Consensus grades fail to consider ambiguous instances where a speech sample may contain multiple emotions, as captured through grader opinion uncertainty. We demonstrate that using the probability density function of the emotion grades as targets instead of the commonly used consensus grades, provide better performance on benchmark evaluation sets compared to results reported in the literature. We show that a saliency driven foundation model (FM) representation selection helps to train a state-of-the-art speech emotion model for both dimensional and categorical emotion recognition. Comparing representations obtained from different FMs, we observed that focusing on overall test-set performance can be deceiving, as it fails to reveal the models generalization capacity across speakers and gender. We demonstrate that performance evaluation across multiple test-sets and performance analysis across gender and speakers are useful in assessing usefulness of emotion models. Finally, we demonstrate that label uncertainty and data-skew pose a challenge to model evaluation, where instead of using the best hypothesis, it is useful to consider the 2- or 3-best hypotheses.
Large language model (LLM)-based test generation has gained attention in software engineering, yet most studies evaluate LLMs' ability to generate unit tests in a single attempt for a given language, missing the opportunity to leverage LLM diversity for more robust testing. This paper introduces PolyTest, a novel approach that enhances test generation by exploiting polyglot and temperature-controlled diversity. PolyTest systematically leverages these properties in two complementary ways: (1) Cross-lingual test generation, where tests are generated in multiple languages at zero temperature and then unified; (2) Diverse test sampling, where multiple test sets are generated within the same language at a higher temperature before unification. A key insight is that LLMs can generate diverse yet contradicting tests -- same input, different expected outputs -- across languages and generations. PolyTest mitigates inconsistencies by unifying test sets, fostering self-consistency and improving overall test quality. Unlike single-language or single-attempt approaches, PolyTest enhances testing without requiring on-the-fly execution, making it particularly beneficial for weaker-performing languages. We evaluate PolyTest on Llama3-70B, GPT-4o, and GPT-3.5 using EvalPlus, generating tests in five languages (Java, C, Python, JavaScript, and a CSV-based format) at temperature 0 and sampling multiple sets at temperature 1. We observe that LLMs frequently generate contradicting tests across settings, and that PolyTest significantly improves test quality across all considered metrics -- number of tests, passing rate, statement/branch coverage (up to +9.01%), and mutation score (up to +11.23%). Finally, PolyTest outperforms Pynguin in test generation, passing rate, and mutation score.
LLM cascades deploy small LLMs to answer most queries, limiting the use of large and expensive LLMs to difficult queries. This approach can significantly reduce costs without impacting performance. However, risk-sensitive domains such as finance or medicine place an additional premium on avoiding model errors. Since even the most expensive models are susceptible to making mistakes, applications in these domains benefit from allowing LLM systems to completely abstain from answering difficult queries. Introducing abstention poses a design question for LLM cascades: should abstention only be allowed at the final model or also at earlier models? Since the error patterns of small and large models are correlated, allowing earlier models to abstain may reduce inference costs and latency by anticipating abstention decisions by expensive and slow models, thus avoiding the need to run these models. We investigate the benefits of such "early abstention" in LLM cascades and find that it reduces overall test loss by 2.2% on average across six benchmarks (GSM8K, MedMCQA, MMLU, TriviaQA, TruthfulQA, and XSum). These gains result from a more effective use of abstention, trading a 4.1% average increase in the overall abstention rate for a 13.0% reduction in cost and a 5.0% reduction in error rate. Our findings demonstrate the possibility of leveraging correlations between the error patterns of different language models to drive performance improvements for LLM systems with abstention.
Hematological disorders, which involve a variety of malignant conditions and genetic diseases affecting blood formation, present significant diagnostic challenges. One such major challenge in clinical settings is differentiating Erythroblast from WBCs. Our approach evaluates the efficacy of various machine learning (ML) classifiers$\unicode{x2014}$SVM, XG-Boost, KNN, and Random Forest$\unicode{x2014}$using the ResNet-50 deep learning model as a backbone in detecting and differentiating erythroblast blood smear images across training splits of different sizes. Our findings indicate that the ResNet50-SVM classifier consistently surpasses other models' overall test accuracy and erythroblast detection accuracy, maintaining high performance even with minimal training data. Even when trained on just 1% (168 images per class for eight classes) of the complete dataset, ML classifiers such as SVM achieved a test accuracy of 86.75% and an erythroblast precision of 98.9%, compared to 82.03% and 98.6% of pre-trained ResNet-50 models without any classifiers. When limited data is available, the proposed approach outperforms traditional deep learning models, thereby offering a solution for achieving higher classification accuracy for small and unique datasets, especially in resource-scarce settings.
During the pretraining phase, large language models (LLMs) acquire vast amounts of knowledge from extensive text corpora. Nevertheless, in later stages such as fine-tuning and inference, the model may encounter knowledge not covered in the initial training, which can lead to hallucinations and degraded performance. This issue has a profound impact on the model's capabilities, as it will inevitably face out-of-scope knowledge after pretraining. Furthermore, fine-tuning is often required to adapt LLMs to domain-specific tasks. However, this phenomenon limits the model's ability to learn and integrate new information during fine-tuning. The effectiveness of fine-tuning largely depends on the type of knowledge involved. Existing research suggests that fine-tuning the model on partially mastered knowledge-for instance, question-answer pairs where the model has a chance of providing correct responses under non-greedy decoding-can enable the model to acquire new knowledge while mitigating hallucination. Notably, this approach can still lead to the forgetting of fully mastered knowledge, constraining the fine-tuning dataset to a narrower range and limiting the model's overall potential for improvement. Given the model's intrinsic reasoning abilities and the interconnectedness of different knowledge areas, it is likely that as the model's capacity to utilize existing knowledge improves during fine-tuning, previously unmastered knowledge may become more understandable. To explore this hypothesis, we conducted experiments and, based on the results, proposed a two-stage fine-tuning strategy. This approach not only improves the model's overall test accuracy and knowledge retention but also preserves its accuracy on previously mastered content. When fine-tuning on the WikiQA dataset, our method increases the amount of knowledge acquired by the model in this stage by 24%.
One of the growing trends in machine learning is the use of data generation techniques, since the performance of machine learning models is dependent on the quantity of the training dataset. However, in many medical applications, collecting large datasets is challenging due to resource constraints, which leads to overfitting and poor generalization. This paper introduces a novel method, Artificial Data Point Generation in Clustered Latent Space (AGCL), designed to enhance classification performance on small medical datasets through synthetic data generation. The AGCL framework involves feature extraction, K-means clustering, cluster evaluation based on a class separation metric, and the generation of synthetic data points from clusters with distinct class representations. This method was applied to Parkinson's disease screening, utilizing facial expression data, and evaluated across multiple machine learning classifiers. Experimental results demonstrate that AGCL significantly improves classification accuracy compared to baseline, GN and kNNMTD. AGCL achieved the highest overall test accuracy of 83.33% and cross-validation accuracy of 90.90% in majority voting over different emotions, confirming its effectiveness in augmenting small datasets.
Neural operators (NOs) employ deep neural networks to learn mappings between infinite-dimensional function spaces. Deep operator network (DeepONet), a popular NO architecture, has demonstrated success in the real-time prediction of complex dynamics across various scientific and engineering applications. In this work, we introduce a random sampling technique to be adopted during the training of DeepONet, aimed at improving the generalization ability of the model, while significantly reducing the computational time. The proposed approach targets the trunk network of the DeepONet model that outputs the basis functions corresponding to the spatiotemporal locations of the bounded domain on which the physical system is defined. While constructing the loss function, DeepONet training traditionally considers a uniform grid of spatiotemporal points at which all the output functions are evaluated for each iteration. This approach leads to a larger batch size, resulting in poor generalization and increased memory demands, due to the limitations of the stochastic gradient descent (SGD) optimizer. The proposed random sampling over the inputs of the trunk net mitigates these challenges, improving generalization and reducing memory requirements during training, resulting in significant computational gains. We validate our hypothesis through three benchmark examples, demonstrating substantial reductions in training time while achieving comparable or lower overall test errors relative to the traditional training approach. Our results indicate that incorporating randomization in the trunk network inputs during training enhances the efficiency and robustness of DeepONet, offering a promising avenue for improving the framework's performance in modeling complex physical systems.
Software testing is a crucial phase in the software development lifecycle (SDLC), ensuring that products meet necessary functional, performance, and quality benchmarks before release. Despite advancements in automation, traditional methods of generating and validating test cases still face significant challenges, including prolonged timelines, human error, incomplete test coverage, and high costs of manual intervention. These limitations often lead to delayed product launches and undetected defects that compromise software quality and user satisfaction. The integration of artificial intelligence (AI) into software testing presents a promising solution to these persistent challenges. AI-driven testing methods automate the creation of comprehensive test cases, dynamically adapt to changes, and leverage machine learning to identify high-risk areas in the codebase. This approach enhances regression testing efficiency while expanding overall test coverage. Furthermore, AI-powered tools enable continuous testing and self-healing test cases, significantly reducing manual oversight and accelerating feedback loops, ultimately leading to faster and more reliable software releases. This paper explores the transformative potential of AI in improving test case generation and validation, focusing on its ability to enhance efficiency, accuracy, and scalability in testing processes. It also addresses key challenges associated with adapting AI for testing, including the need for high quality training data, ensuring model transparency, and maintaining a balance between automation and human oversight. Through case studies and examples of real-world applications, this paper illustrates how AI can significantly enhance testing efficiency across both legacy and modern software systems.
While data augmentation (DA) is generally applied to input data, several studies have reported that applying DA to hidden layers in neural networks, i.e., feature augmentation, can improve performance. However, in previous studies, the layers to which DA is applied have not been carefully considered, often being applied randomly and uniformly or only to a specific layer, leaving room for arbitrariness. Thus, in this study, we investigated the trends of suitable layers for applying DA in various experimental configurations, e.g., training from scratch, transfer learning, various dataset settings, and different models. In addition, to adjust the suitable layers for DA automatically, we propose the adaptive layer selection (AdaLASE) method, which updates the ratio to perform DA for each layer based on the gradient descent method during training. The experimental results obtained on several image classification datasets indicate that the proposed AdaLASE method altered the ratio as expected and achieved high overall test accuracy.
Vision-Language Models (VLMs) such as CLIP have yielded unprecedented performance for zero-shot image classification, yet their generalization capability may still be seriously challenged when confronted to domain shifts. In response, we present Weight Average Test-Time Adaptation (WATT) of CLIP, a pioneering approach facilitating full test-time adaptation (TTA) of this VLM. Our method employs a diverse set of templates for text prompts, augmenting the existing framework of CLIP. Predictions are utilized as pseudo labels for model updates, followed by weight averaging to consolidate the learned information globally. Furthermore, we introduce a text ensemble strategy, enhancing overall test performance by aggregating diverse textual cues. Our findings underscore the efficacy of WATT in enhancing performance across diverse datasets, including CIFAR-10-C, CIFAR-10.1, CIFAR-100-C, VisDA-C, and several other challenging datasets, effectively covering a wide range of domain shifts. Notably, these enhancements are achieved without necessitating additional model transformations or trainable modules. Moreover, compared to other Test-Time Adaptation methods, our approach can operate effectively with just a single image. Highlighting the potential of innovative test-time strategies, this research emphasizes their role in fortifying the adaptability of VLMs. The implementation is available at: \url{https://github.com/Mehrdad-Noori/WATT.git}.
We consider the two-sample testing problem for networks, where the goal is to determine whether two sets of networks originated from the same stochastic model. Assuming no vertex correspondence and allowing for different numbers of nodes, we address a fundamental network testing problem that goes beyond simple adjacency matrix comparisons. We adopt the stochastic block model (SBM) for network distributions, due to their interpretability and the potential to approximate more general models. The lack of meaningful node labels and vertex correspondence translate to a graph matching challenge when developing a test for SBMs. We introduce an efficient algorithm to match estimated network parameters, allowing us to properly combine and contrast information within and across samples, leading to a powerful test. We show that the matching algorithm, and the overall test are consistent, under mild conditions on the sparsity of the networks and the sample sizes, and derive a chi-squared asymptotic null distribution for the test. Through a mixture of theoretical insights and empirical validations, including experiments with both synthetic and real-world data, this study advances robust statistical inference for complex network data.
Achieving accurate human identification through RF imaging has been a persistent challenge, primarily attributed to the limited aperture size and its consequent impact on imaging resolution. The existing imaging solution enables tasks such as pose estimation, activity recognition, and human tracking based on deep neural networks by estimating skeleton joints. In contrast to estimating joints, this paper proposes to improve imaging resolution by estimating the human figure as a whole using conditional generative adversarial networks (cGAN). In order to reduce training complexity, we use an estimated spatial spectrum using the MUltiple SIgnal Classification (MUSIC) algorithm as input to the cGAN. Our system generates environmentally independent, high-resolution images that can extract unique physical features useful for human identification. We use a simple convolution layers-based classification network to obtain the final identification result. From the experimental results, we show that resolution of the image produced by our trained generator is high enough to enable human identification. Our finding indicates high-resolution accuracy with 5% mean silhouette difference to the Kinect device. Extensive experiments in different environments on multiple testers demonstrate that our system can achieve 93% overall test accuracy in unseen environments for static human target identification.
Large Language Models (LLMs) prompted to generate chain-of-thought (CoT) exhibit impressive reasoning capabilities. Recent attempts at prompt decomposition toward solving complex, multi-step reasoning problems depend on the ability of the LLM to simultaneously decompose and solve the problem. A significant disadvantage is that foundational LLMs are typically not available for fine-tuning, making adaptation computationally prohibitive. We believe (and demonstrate) that problem decomposition and solution generation are distinct capabilites, better addressed in separate modules, than by one monolithic LLM. We introduce DaSLaM, which uses a decomposition generator to decompose complex problems into subproblems that require fewer reasoning steps. These subproblems are answered by a solver. We use a relatively small (13B parameters) LM as the decomposition generator, which we train using policy gradient optimization to interact with a solver LM (regarded as black-box) and guide it through subproblems, thereby rendering our method solver-agnostic. Evaluation on multiple different reasoning datasets reveal that with our method, a 175 billion parameter LM (text-davinci-003) can produce competitive or even better performance, compared to its orders-of-magnitude larger successor, GPT-4. Additionally, we show that DaSLaM is not limited by the solver's capabilities as a function of scale; e.g., solver LMs with diverse sizes give significant performance improvement with our solver-agnostic decomposition technique. Exhaustive ablation studies evince the superiority of our modular finetuning technique over exorbitantly large decomposer LLMs, based on prompting alone.
We consider availability data poisoning attacks, where an adversary aims to degrade the overall test accuracy of a machine learning model by crafting small perturbations to its training data. Existing poisoning strategies can achieve the attack goal but assume the victim to employ the same learning method as what the adversary uses to mount the attack. In this paper, we argue that this assumption is strong, since the victim may choose any learning algorithm to train the model as long as it can achieve some targeted performance on clean data. Empirically, we observe a large decrease in the effectiveness of prior poisoning attacks if the victim employs an alternative learning algorithm. To enhance the attack transferability, we propose Transferable Poisoning, which first leverages the intrinsic characteristics of alignment and uniformity to enable better unlearnability within contrastive learning, and then iteratively utilizes the gradient information from supervised and unsupervised contrastive learning paradigms to generate the poisoning perturbations. Through extensive experiments on image benchmarks, we show that our transferable poisoning attack can produce poisoned samples with significantly improved transferability, not only applicable to the two learners used to devise the attack but also to learning algorithms and even paradigms beyond.
To push forward the important emerging research field surrounding multi-area recurrent neural networks (RNNs), we expand theoretically and empirically on the provably stable RNNs of RNNs introduced by Kozachkov et al. in "RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks". We prove relaxed stability conditions for salient special cases of this architecture, most notably for a global workspace modular structure. We then demonstrate empirical success for Global Workspace Sparse Combo Nets with a small number of trainable parameters, not only through strong overall test performance but also greater resilience to removal of individual subnetworks. These empirical results for the global workspace inter-area topology are contingent on stability preservation, highlighting the relevance of our theoretical work for enabling modular RNN success. Further, by exploring sparsity in the connectivity structure between different subnetwork modules more broadly, we improve the state of the art performance for stable RNNs on benchmark sequence processing tasks, thus underscoring the general utility of specialized graph structures for multi-area RNNs.
Even when aggregate accuracy is high, state-of-the-art NLP models often fail systematically on specific subgroups of data, resulting in unfair outcomes and eroding user trust. Additional data collection may not help in addressing these weaknesses, as such challenging subgroups may be unknown to users, and underrepresented in the existing and new data. We propose Targeted Data Generation (TDG), a framework that automatically identifies challenging subgroups, and generates new data for those subgroups using large language models (LLMs) with a human in the loop. TDG estimates the expected benefit and potential harm of data augmentation for each subgroup, and selects the ones most likely to improve within group performance without hurting overall performance. In our experiments, TDG significantly improves the accuracy on challenging subgroups for state-of-the-art sentiment analysis and natural language inference models, while also improving overall test accuracy.
The performance of large language models (LLMs) on existing reasoning benchmarks has significantly improved over the past years. In response, we present JEEBench, a considerably more challenging benchmark dataset for evaluating the problem solving abilities of LLMs. We curate 515 challenging pre-engineering mathematics, physics and chemistry problems from the highly competitive IIT JEE-Advanced exam. Long-horizon reasoning on top of deep in-domain knowledge is essential for solving problems in this benchmark. Our evaluation on various open-source and proprietary models reveals that the highest performance, even after using techniques like self-consistency, self-refinement and chain-of-thought prompting, is less than 40%. The typical failure modes of GPT-4, the best model, are errors in algebraic manipulation, difficulty in grounding abstract concepts into mathematical equations accurately and failure in retrieving relevant domain-specific concepts. We also observe that by mere prompting, GPT-4 is unable to assess risk introduced by negative marking for incorrect answers. For this, we develop a post-hoc confidence-thresholding method over self-consistency, which enables effective response selection. We hope that our challenging benchmark will guide future re-search in problem-solving using LLMs.