AI is used in image classification to automatically categorize and label images based on their content. Through deep learning algorithms, neural networks can learn to recognize patterns, objects, and features in images, enabling applications such as facial recognition, object detection, and automated image tagging.
Researchers at Princeton and Stanford have developed CALDERA, a novel algorithm to compress large language models (LLMs) by leveraging low-rank and low-precision techniques. This innovation reduces energy use, cost, and storage needs, enabling LLM deployment on consumer devices without sacrificing performance.
Researchers at the University of Tokyo developed ADOPT, a novel optimization algorithm that overcomes convergence issues in adaptive gradient methods, promising more reliable and efficient training for deep learning models.
Researchers unveil PITOME, a novel token merging technique that accelerates transformer models while preserving critical information for enhanced AI efficiency.
Researchers developed a multiclass AI model using EfficientNet and KAN to detect Wolfgang Beltracchi's art forgeries, revealing key stylistic traits and improving forgery detection techniques.
A new study introduces the YesBut dataset, designed to evaluate how well vision-language models comprehend satire, highlighting significant gaps in current model capabilities.
Researchers developed the "Deepdive" dataset and benchmarked deep learning models to automate the classification of deep-sea biota in the Great Barrier Reef, achieving significant accuracy with the Inception-ResNet model.
The novel SBDet model introduces a relaxed rotation-equivariant network (R2Net) that improves object detection in scenarios with symmetry-breaking or non-rigid transformations. This innovation offers greater accuracy and robustness in real-world visual tasks like autonomous driving and geosciences.
Researchers introduced an adaptive backdoor attack method to steal private data from pre-trained large language models (LLMs). This method, tested on models like GPT-3.5-turbo, achieved a 92.5% success rate. By injecting triggers during model customization and activating them during inference, attackers can extract sensitive information, underscoring the need for advanced security measures.
Researchers at Meta Research introduced Hallucinating Datasets with Evolution Strategies (HaDES), a novel method for dataset distillation in reinforcement learning (RL). HaDES compresses extensive datasets into a few synthetic examples, enhancing the training efficiency of RL models by integrating behavior distillation to optimize state-action pairs for expert policy training, demonstrating superior performance across multiple environments.
The European project SIGNIFICANCE, using AI and deep learning, developed a platform to combat the illegal trafficking of cultural heritage goods. By identifying, tracking, and blocking illegal online activities, the platform increased the detection of illegal artifacts by 10-15%, aiding law enforcement in safeguarding cultural heritage.
Researchers introduced a semi-supervised concept bottleneck model (SSCBM) that enhances concept prediction accuracy and interpretability by using pseudo-labels and alignment loss with both labeled and unlabeled data. The SSCBM framework demonstrated high effectiveness, achieving superior performance with only 20% labeled data compared to fully supervised settings.
Researchers provide an introductory guide to vision-language models, detailing their functionalities, training methods, and evaluation processes. The study emphasizes the potential and challenges of integrating visual data with language models to advance AI applications.
Researchers harness convolutional neural networks (CNNs) to recognize Shen embroidery, achieving 98.45% accuracy. By employing transfer learning and enhancing MobileNet V1 with spatial pyramid pooling, they provide crucial technical support for safeguarding this cultural art form.
Researchers introduced a multi-stage progressive detection method utilizing a Swin transformer to accurately identify water deficit in vertical greenery plants. By integrating classification, semantic segmentation, and object detection, the approach significantly improved detection accuracy compared to traditional methods like R-CNN and YOLO, offering promising solutions for urban greenery management.
Researchers introduced a deep convolutional neural network (DCNN) model for accurately detecting and classifying grape leaf diseases. Leveraging a dataset of grape leaf images, the DCNN model outperformed conventional CNN models, demonstrating superior accuracy and reliability in identifying black rot, ESCA, leaf blight, and healthy specimens.
Researchers integrated gradient quantization (GQ) into DenseNet architecture to improve image recognition (IR). By optimizing feature reuse and introducing GQ for parallel training, they achieved superior accuracy and accelerated training speed, overcoming communication bottlenecks.
This study in Nature explores the application of convolutional neural networks (CNNs) in classifying infrared (IR) images for concealed object detection in security scanning. Leveraging a ResNet-50 model and transfer learning, the researchers refined pre-processing techniques such as k-means and fuzzy-c clustering to improve classification accuracy.
Researchers introduced OCTDL, an open-access dataset comprising over 2000 labeled OCT images of retinal diseases, including AMD, DME, and others. Utilizing high-resolution OCT scans obtained from an Optovue Avanti RTVue XR system, the dataset facilitated the development of deep learning models for disease classification. Validation with VGG16 and ResNet50 architectures demonstrated high performance, indicating OCTDL's potential for advancing automatic processing and early disease detection in ophthalmology.
Recent research in few-shot fine-grained image classification (FSFGIC) has seen the development of various methods, including class representation learning and global/local deep feature representation techniques. These advancements aim to improve generalization, overcome distribution biases, and enhance discriminative feature representation, yet challenges such as overfitting and efficiency persist, necessitating further investigation.
This paper presents the groundbreaking lifelong learning optical neural network (L2ONN), offering efficient and scalable AI systems through photonic computing. L2ONN's innovative architecture harnesses sparse photonic connections and parallel processing, surpassing traditional electronic models in efficiency, capacity, and lifelong learning capabilities, with implications for various applications from vision classification to medical diagnosis.
Terms
While we only use edited and approved content for Azthena
answers, it may on occasions provide incorrect responses.
Please confirm any data provided with the related suppliers or
authors. We do not provide medical advice, if you search for
medical information you must always consult a medical
professional before acting on any information provided.
Your questions, but not your email details will be shared with
OpenAI and retained for 30 days in accordance with their
privacy principles.
Please do not ask questions that use sensitive or confidential
information.
Read the full Terms & Conditions.