A5033198334- Advanced Neural Network Applications
- Image Retrieval and Classification Techniques
- Medical Image Segmentation Techniques
- Natural Language Processing Techniques
- Cutaneous Melanoma Detection and Management
- Advanced Image and Video Retrieval Techniques
- Domain Adaptation and Few-Shot Learning
- Topic Modeling
- Speech Recognition and Synthesis
- Skin Protection and Aging
- Image and Signal Denoising Methods
- Body Image and Dysmorphia Studies
- Face recognition and analysis
- Generative Adversarial Networks and Image Synthesis
- Visual Attention and Saliency Detection
- Multimodal Machine Learning Applications
- Advancements in Photolithography Techniques
- Artificial Intelligence in Healthcare
- Imbalanced Data Classification Techniques
- Radiomics and Machine Learning in Medical Imaging
- Advanced Image Processing Techniques
- Age of Information Optimization
- COVID-19 diagnosis using AI
- Advanced Data Compression Techniques
- Explainable Artificial Intelligence (XAI)
Northeastern University
2022-2024
Universidad del Noreste
2023
With the ever-increasing popularity of edge devices, it is necessary to implement real-time segmentation on for autonomous driving and many other applications. Vision Transformers (ViTs) have shown considerably stronger results vision tasks. However, ViTs with fullattention mechanism usually consume a large number computational resources, leading difficulties real- time inference devices. In this paper, we aim derive fewer computations fast speed facilitate dense prediction semantic To...
Structured pruning for large language models (LLMs) has garnered significant academic interest due to its ability efficiently compress and accelerate LLMs by eliminating redundant weight groups at a coarse-grained granularity. Current structured methods typically depend on singular granularity assessing importance, resulting in notable performance degradation downstream tasks. Intriguingly, our empirical investigations reveal that utilizing unstructured pruning, which achieves better...
The research in real-time segmentation mainly focuses on desktop GPUs. However, autonomous driving and many other applications rely the edge, current arts are far from goal. In addition, recent advances vision transformers also inspire us to re-design network architecture for dense prediction task. this work, we propose combine self attention block with lightweight convolutions form new building blocks, employ latency constraints search an efficient sub-network. We train MLP model based...
Gliomas, often known as low-grade gliomas, are malignant brain tumors. Codeletion of chromosomal arms 1p/19q has been connected with a good response to treatment in gliomas (LGG) several studies. For planning, the ability anticipate 1p19q status is crucial. This research’s purpose develop noninvasive approach based on MR images using our efficient CNNs. While public networks like VGGNet, GoogleNet, and other well-known can use transfer learning identify cancer MRI, model contains large...
The interpretability of deep learning models has emerged as a compelling area in artificial intelligence research. safety criteria for medical imaging are highly stringent, and required an explanation. However, existing convolutional neural network solutions left ventricular segmentation viewed terms inputs outputs. Thus, the CNNs come into spotlight. Since data limited, many methods to fine-tune that popular transfer have been built using massive public ImageNet datasets by method....
Despite the remarkable strides of Large Language Models (LLMs) in various fields, wide applications LLMs on edge devices are limited due to their massive parameters and computations. To address this, quantization is commonly adopted generate lightweight with efficient computations fast inference. However, Post-Training Quantization (PTQ) methods dramatically degrade quality when quantizing weights, activations, KV cache together below 8 bits. Besides, many Quantization-Aware Training (QAT)...
Class Incremental Learning (CIL) is challenging due to catastrophic forgetting. On top of that, Exemplar-free even more forbidden access previous task data. Recent exemplar-free CIL methods attempt mitigate forgetting by synthesizing However, they fail overcome the inability deal with significant domain gap between real and synthetic To these issues, we propose a novel method. Our method adopts multi-distribution matching (MDM) diffusion models unify quality bridge gaps among all domains...
Recent advances in image editing have been driven by the development of denoising diffusion models, marking a significant leap forward this field. Despite these advances, generalization capabilities recent approaches remain constrained. In response to challenge, our study introduces novel framework with enhanced robustness boosting in-context learning capability and unifying language instruction. This incorporates module specifically optimized for tasks, leveraging VMamba Block an...
Large language models (LLMs) have become crucial for many generative downstream tasks, leading to an inevitable trend and significant challenge deploy them efficiently on resource-constrained devices. Structured pruning is a widely used method address this challenge. However, when dealing with the complex structure of multiple decoder layers, general methods often employ common estimation approaches pruning. These lead decline in accuracy specific tasks. In paper, we introduce simple yet...
The rapid progress in artificial intelligence-generated content (AIGC), especially with diffusion models, has significantly advanced development of high-quality video generation. However, current models exhibit demanding computational requirements and high peak memory usage, for generating longer higher-resolution videos. These limitations greatly hinder the practical application on standard hardware platforms. To tackle this issue, we present a novel, training-free framework named...
As edge devices become readily available and indispensable, there is an urgent need for effective efficient intelligent applications to be deployed widespread. However, fairness has always been issue, especially in medical applications. Compared convolutional neuron networks (CNNs), Vision Transformer (ViT) a better ability extract global information, which will contribute alleviating the unfairness problem. Typically, ViTs consume large amounts of computational memory resources, hinders...
As edge devices become readily available and indispensable, there is an urgent need for effective efficient intelligent applications to be deployed widespread. However, fairness has always been issue, especially in medical applications. Although many approaches have proposed mitigate the unfairness problem, their performance not desirable. By examining of different network architectures, we observed that compared pure convolutional neuron (CNN) architecture, hybrid models with CNN Vision...