A5079840947- Topic Modeling
- Natural Language Processing Techniques
- Video Surveillance and Tracking Methods
- Coal Properties and Utilization
- Advanced Image and Video Retrieval Techniques
- Machine Learning and Data Classification
- Face and Expression Recognition
- Geoscience and Mining Technology
- Advanced Graph Neural Networks
- Rock Mechanics and Modeling
- Domain Adaptation and Few-Shot Learning
- Machine Learning and Algorithms
- Multimodal Machine Learning Applications
- Image Enhancement Techniques
- Data Quality and Management
- Hydrocarbon exploration and reservoir analysis
- Human Pose and Action Recognition
- Landslides and related hazards
- Soil and Land Suitability Analysis
- Geochemistry and Geologic Mapping
- Anomaly Detection Techniques and Applications
- Remote-Sensing Image Classification
- Image Retrieval and Classification Techniques
- Advanced Vision and Imaging
- Multi-Criteria Decision Making
Kharazmi University
2013-2024
Amazon (United Kingdom)
2021-2023
Korea Advanced Institute of Science and Technology
2023
Amazon (Germany)
2020-2023
Kootenay Association for Science & Technology
2023
Amazon (United States)
2021
University of Shahrood
2013-2020
BenevolentAI (United Kingdom)
2020
Oxford Brookes University
2011-2013
Graz University of Technology
2006-2010
Adaptive tracking-by-detection methods are widely used in computer vision for tracking arbitrary objects. Current approaches treat the problem as a classification task and use online learning techniques to update object model. However, these updates happen one needs convert estimated position into set of labelled training examples, it is not clear how best perform this intermediate step. Furthermore, objective classifier (label prediction) explicitly coupled tracker (accurate estimation...
Adaptive tracking-by-detection methods are widely used in computer vision for tracking arbitrary objects. Current approaches treat the problem as a classification task and use online learning techniques to update object model. However, these updates happen one needs convert estimated position into set of labelled training examples, it is not clear how best perform this intermediate step. Furthermore, objective classifier (label prediction) explicitly coupled tracker (estimation position). In...
The Visual Object Tracking challenge 2015, VOT2015, aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance. Results 62 are presented. number tested makes VOT 2015 the largest benchmark on tracking to date. For each participating tracker, a short description is provided in appendix. Features VOT2015 go beyond its VOT2014 predecessor are: (i) new dataset twice as large with full annotation targets by rotated bounding boxes and...
Random Forests (RFs) are frequently used in many computer vision and machine learning applications. Their popularity is mainly driven by their high computational efficiency during both training evaluation while achieving state-of-the-art results. However, most applications RFs off-line. This limits usability for practical problems, instance, when data arrives sequentially or the underlying distribution continuously changing. In this paper, we propose a novel on-line random forest algorithm....
Tracking-by-detection is increasingly popular in order to tackle the visual tracking problem. Existing adaptive methods suffer from drifting problem, since they rely on self-updates of an on-line learning method. In contrast previous work that tackled this problem by employing semi-supervised or multiple-instance learning, we show augmenting method with complementary approaches can lead more stable results. particular, use a simple template model as non-adaptive and thus component, novel...
Efficient keypoint-based object detection methods are used in many real-time computer vision applications. These approaches often model an as a collection of keypoints and associated descriptors, then involves first constructing set correspondences between image via descriptor matching, subsequently using these input to robust geometric estimation algorithm such RANSAC find the transformation image. In approaches, is generally constructed offline, does not adapt given environment at runtime....
The Thermal Infrared Visual Object Tracking challenge 2015, VOT-TIR2015, aims at comparing short-term single-object visual trackers that work on thermal infrared (TIR) sequences and do not apply pre-learned models of object appearance. VOT-TIR2015 is the first benchmark tracking in TIR sequences. Results 24 are presented. For each participating tracker, a short description provided appendix. based VOT2013 challenge, but introduces following novelties: (i) newly collected LTIR (Link -- ping...
Large Language Models (LLMs) are capable of performing zero-shot closed-book question answering tasks, based on their internal knowledge stored in parameters during pre-training. However, such internalized might be insufficient and incorrect, which could lead LLMs to generate factually wrong answers. Furthermore, fine-tuning update is expensive. To this end, we propose augment the directly input LLMs. Specifically, first retrieve relevant facts from graph semantic similarities between its...
Random Forests (RFs) have become commonplace in many computer vision applications. Their popularity is mainly driven by their high computational efficiency during both training and evaluation while still being able to achieve state-of-the-art accuracy. This work extends the usage of Semi-Supervised Learning (SSL) problems. We show that traditional decision trees are optimizing multi-class margin maximizing loss functions. From this intuition, we develop a novel definition for unlabeled data,...
A recent dominating trend in tracking called tracking-by-detection uses on-line classifiers order to redetect objects over succeeding frames. Although these methods usually deliver excellent results and run real-time they also tend drift case of wrong updates during the self-learning process. Recent approaches tackled this problem by formulating as either one-shot semi-supervised learning or multiple instance learning. Semi-supervised allows for incorporating priors is more robust occlusions...
Online boosting is one of the most successful online learning algorithms in computer vision. While many challenging problems are inherently multi-class, and its variants only able to solve binary tasks. In this paper, we present Multi-Class LPBoost (OMCLP) which directly applicable multi-class problems. From a theoretical point view, our algorithm tries maximize soft-margin samples. order LP problem settings, perform an efficient variant convex programming, based on primal-dual gradient...
End-to-end neural data-to-text (D2T) generation has recently emerged as an alternative to pipeline-based architectures. However, it faced challenges generalizing new domains and generating semantically consistent text. In this work, we present DataTuner, a neural, end-to-end system that makes minimal assumptions about the data representation target domain. We take two-stage generation-reranking approach, combining fine-tuned language model with semantic fidelity classifier. Each component is...
On-line boosting is one of the most successful on-line algorithms and thus applied in many computer vision applications. However, even though boosting, general, well known to be susceptible class-label noise, mostly self-learning applications such as visual object tracking, where label-noise an inherent problem. This paper studies robustness boosting. Since mainly loss function determines behavior we propose version GradientBoost, which allows us plug arbitrary loss-functions into learner....
This paper introduces a novel classification method termed Alternating Decision Forests (ADFs), which formulates the training of Random explicitly as global loss minimization problem. During training, losses are minimized via keeping an adaptive weight distribution over samples, similar to Boosting methods. In order keep flexible and general possible, we adopt principle employing gradient descent in function space, allows minimize arbitrary losses. Contrary Boosted Trees, our is inherent...
Generating novel molecules with optimal properties is a crucial step in many industries such as drug discovery. Recently, deep generative models have shown promising way of performing de-novo molecular design. Although graph are currently available they either size dependency their number parameters, limiting use to only very small graphs or formulated sequence discrete actions needed construct graph, making the output non-differentiable w.r.t model therefore preventing them be used...
Large Language Models (LLMs) are capable of performing zero-shot closed-book question answering tasks, based on their internal knowledge stored in parameters during pre-training. However, such internalized might be insufficient and incorrect, which could lead LLMs to generate factually wrong answers. Furthermore, fine-tuning update is expensive. To this end, we propose augment the directly input LLMs. Specifically, first retrieve relevant facts from graph semantic similarities between its...
In this work, we provide a new formulation for Graph Convolutional Neural Networks (GCNNs) link prediction on graph data that addresses common challenges biomedical knowledge graphs (KGs). We introduce regularized attention mechanism to GCNNs not only improves performance clean datasets, but also favorably accommodates noise in KGs, pervasive issue real-world applications. Further, explore visualization methods interpretable modelling and illustrate how the learned representation can be...
While models have reached superhuman performance on popular question answering (QA) datasets such as SQuAD, they yet to outperform humans the task of itself. In this paper, we investigate if are learning reading comprehension from QA by evaluating BERT-based across five datasets. We evaluate their generalizability out-of-domain examples, responses missing or incorrect data, and ability handle variations. find that no single dataset is robust all our experiments identify shortcomings in both...
We introduce Mintaka, a complex, natural, and multilingual dataset designed for experimenting with end-to-end question-answering models. Mintaka is composed of 20,000 question-answer pairs collected in English, annotated Wikidata entities, translated into Arabic, French, German, Hindi, Italian, Japanese, Portuguese, Spanish total 180,000 samples. includes 8 types complex questions, including superlative, intersection, multi-hop which were naturally elicited from crowd workers. run baselines...