A5100772524- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Geochemistry and Geologic Mapping
- Geology and Paleoclimatology Research
- COVID-19 diagnosis using AI
- Anomaly Detection Techniques and Applications
- Natural Language Processing Techniques
- Neural Networks and Applications
- Botany, Ecology, and Taxonomy Studies
- Cancer-related molecular mechanisms research
- Machine Learning and ELM
- Advanced Neural Network Applications
- Data Stream Mining Techniques
- Wildlife Ecology and Conservation
- Geological Modeling and Analysis
- Human Pose and Action Recognition
- Privacy-Preserving Technologies in Data
- Logic, programming, and type systems
- Topic Modeling
- Human-Animal Interaction Studies
- Bayesian Modeling and Causal Inference
- Geological and Geochemical Analysis
- Health and Medical Research Impacts
- Adversarial Robustness in Machine Learning
- Species Distribution and Climate Change
Flinders University
2023-2025
IBM (United States)
2023-2024
Samsung (United States)
2024
Research!America (United States)
2024
Bush Heritage Australia
2024
Georgia Institute of Technology
2019-2023
K Lab (United States)
2023
San Francisco State University
2023
Centre for Remote Health
2023
Northeastern University
2021
Computer vision models suffer from a phenomenon known as catastrophic forgetting when learning novel concepts continuously shifting training data. Typical solutions for this continual problem require extensive rehearsal of previously seen data, which increases memory costs and may violate data privacy. Recently, the emergence large-scale pre-trained transformer has enabled prompting approaches an alternative to data-rehearsal. These rely on key-query mechanism generate prompts have been...
Modern computer vision applications suffer from catastrophic forgetting when incrementally learning new concepts over time. The most successful approaches to alleviate this require extensive replay of previously seen data, which is problematic memory constraints or data legality concerns exist. In work, we consider the high-impact problem Data-Free Class-Incremental Learning (DFCIL), where an incremental agent must learn time without storing generators training past tasks. One approach for...
Difficult experiments in training neural networks often fail to converge due what is known as the flat-spot problem, where gradient of hidden neurons network diminishes value, rending weight update process ineffective. Whereas a first-order algorithm can address this issue by learning parameters normalize neuron activations, second-order algorithms cannot afford additional given that they include large Jacobian matrix calculation. This paper proposes Levenberg-Marquardt with compression...
Continual learning is a setting where machine models learn novel concepts from continuously shifting training data, while simultaneously avoiding degradation of knowledge on previously seen classes which may disappear the data for extended periods time (a phenomenon known as catastrophic forgetting problem). Current approaches continual single expanding task (aka class-incremental learning) require extensive rehearsal to avoid this knowledge. Unfortunately, comes at cost memory, and it also...
Abstract We examine a recently developed physics-based tropical cyclone rainfall (TCR) model and apply it to assess the mechanisms that dominate magnitude spatial distribution of TC rainfall, with Hurricanes Isabel (2003) Irene (2011) as study cases. evaluate TCR using Weather Research Forecasting (WRF) Model simulations. TCR-generated fields for two storms compare well WRF estimates in terms both azimuthal mean distributions. When coupled hydrologic model, generates flood peaks over...
Large-scale pre-trained Vision & Language (VL) models have shown remarkable performance in many applications, enabling replacing a fixed set of supported classes with zero-shot open vocabulary reasoning over (almost arbitrary) natural language prompts. However, recent works uncovered fundamental weakness these models. For example, their difficulty to understand Visual Concepts (VLC) that go 'beyond nouns' such as the meaning non-object words (e.g., attributes, actions, relations, states,...
Recent works demonstrate a remarkable ability to customize text-to-image diffusion models while only providing few example images. What happens if you try such using multiple, fine-grained concepts in sequential (i.e., continual) manner? In our work, we show that recent state-of-the-art customization of suffer from catastrophic forgetting when new arrive sequentially. Specifically, adding concept, the generate high quality images past, similar degrade. To circumvent this forgetting, propose...
This paper presents a novel optimization method used to design thin-wire antennas approximate any arbitrary antenna gain pattern. These types of designs may be useful for specific tracking-search radars or telecommunication systems trying maximize the footprint without significant sidelobe power loss. A genetic algorithm (GA) is optimize match predefined Two different 3-D topologies are allowed. The first conventional bent long wire crooked and second branching tree antenna. resembles...
Sulfur-sulfur bonds are ubiquitous across broad classes of natural products, peptides and proteins, drug molecules, synthetic polymers materials. The ability to make break these in a controlled manner is critical for their many scientific technological applications. In this study, we report the discovery new unusual S-S metathesis reaction linear organic trisulfides. When exposed certain polar aprotic solvents, trisulfides were found undergo spontaneous metathesis, with equilibrium...
To determine ambulance transport rates and investigate predictors for use by patients with acute myocardial infarction (AMI) in Australia.A prospective, cross-sectional descriptive survey using structured interviews. It included who were admitted to two hospitals (Western, Bendigo, Melbourne, Victoria, Australia) AMI between 1 October 2004 31 March 2005, data collected semistructured interview medical record review. Data analysed statistics, univariate multivariate analysis SPSS.105...
Rehearsal is a critical component for class-incremental continual learning, yet it requires substantial memory budget. Our work investigates whether we can significantly reduce this budget by leveraging unlabeled data from an agent's environment in realistic and challenging learning paradigm. Specifically, explore formalize novel semi-supervised (SSCL) setting, where labeled scarce non-i.i.d. the plentiful. Importantly, distributions SSCL setting are therefore reflect object class...
The positive impact of evidence-based practice on health service performance and outcomes is well described.1-3 Marita Titler, an expert in from the United States, has observed that "only by putting into what learned research will care be made safer".4 However, consistent widespread application largely dependent a service's culture capacity its staff.5 It therefore important to build capacity, as this improves ability clinicians apply new knowledge improve outcomes.5 In addition, strong been...
Journal Article Practicing New Historicism Get access Historicism. By Catherine Gallagher and Stephen Greenblatt. 1993. Chicago: U of Chicago P, 2000. 249 pp. Cloth $30.00. Jon Smith Mississippi State University Search for other works by this author on: Oxford Academic Google Scholar ISLE: Interdisciplinary Studies in Literature Environment, Volume 8, Issue 2, Summer 2001, Pages 284–285, https://doi.org/10.1093/isle/8.2.284 Published: 01 July 2001
SurveyCover.Redoubt, a massive snow-and ice-covered 10,200-ft-high volcano located 105 miles southwest of Anchorage, has erupted three times in the past 100 years (1902,.The last eruption was characterized by repeated episodes lava-dome growth and subsequent destruction.The resulting pyroclastic flows mixed with ice snow to create lahars, several which completely inundated 1-mile-wide Drift River valley flowed 25 enter sea.The largest flows-comparable average discharge Mississippi...
Recently, large-scale pre-trained Vision-and-Language (VL) foundation models have demonstrated remarkable capabilities in many zero-shot downstream tasks, achieving competitive results for recognizing objects defined by as little short text prompts. However, it has also been shown that VL are still brittle Structured Concept (SVLC) reasoning, such the ability to recognize object attributes, states, and inter-object relations. This leads reasoning mistakes, which need be corrected they occur...
Abstract Timely and accurate detection identification of species are crucial for monitoring wildlife conservation management. Technological advances, including connectivity camera traps to mobile phone networks artificial intelligence (AI) algorithms automated identification, can potentially improve the timeliness accuracy identification. Adoption this new technology, however, is often seen as cost-prohibitive it has been difficult calculate cost savings or qualitative benefits over life...
We first pose the Unsupervised Progressive Learning (UPL) problem: an online representation learning problem in which learner observes a non-stationary and unlabeled data stream, growing number of features that persist over time even though is not stored or replayed. To solve UPL we propose Self-Taught Associative Memory (STAM) architecture. Layered hierarchies STAM modules learn based on combination clustering, novelty detection, forgetting outliers, storing only prototypical rather than...
Knowledge distillation (KD) is a substantial strategy for transferring learned knowledge from one neural network model to another. A vast number of methods have been developed this strategy. While most method designs more efficient way facilitate transfer, less attention has put on comparing the effect sources such as features, logits, and gradients. This work provides new perspective motivate set strategies by approximating classical KL-divergence criteria with different sources, making...
Abstract A geostatistical method to quantify the small-scale 3D–time structure of drop size distribution (DSD) from ground level up melting layer using radar and disdrometer data is presented. First, reflectivity fields are used estimate large-scale properties a rain event, such as apparent motion, spatial anisotropy, temporal innovation. The retrieved quantities then combined with independent time series variogram each DSD parameter. key point in procedure use new metric for measuring...
There are no conventional lymphatics in the brain but physiological studies have revealed a substantial and immunologically significant lymphatic drainage from to cervical lymph nodes.CSF drains via cribriform plate nasal mucosa nodes rats sheep lesser extent humans.More for range of human neurological disorders is interstitial fluid (ISF) solutes parenchyma along capillary artery walls.Tracers injected into grey matter drain basement membranes walls capillaries cerebral arteries.Lymphatic...