Okay, so here we have our lightning talk session, and if we could then move over, let's see, three slides, to get to lightning talk presenter number three. All right, perfect. And then we'll just iterate through as each presenter comes up from here on out. Thank you. Hello everyone, I'm Fardin. I'm from the business school here at UT. So what I've done is, we have done a deep learning based diagnostic framework for colorectal cancer using histopathological images. So colorectal cancer is one of the deadliest cancers, and globally, an early detection is the key for survival. So what we have done, like I've done, is utilized the LC25000 data set and developed a lightweight diagnostic model using MobileNet V2. So it's a deep learning architecture, and we use a custom preprocessing pipeline that includes grayscale conversion, noise reduction, image enhancement, and counter -based cropping to prepare high-quality input for the model. So, what was the result? The result was it achieved 99.95% accuracy and perfect scores in precision, recall, and F1 score for both benign and malignant tissue. This performance not only demonstrates high reliability, but also makes the model highly suitable for resource-constrained environments like rural settings, and we believe this framework can play a vital role in scalable AI -driven diagnostic systems and improve early detection globally. Thank you. Good morning, everyone. I'm Gokul Rahman. I'm a master student at University of Deliver. So emergent droplets are central to pharmaceutics and drug delivery application. And to optimize this application, we have to look into their behavior, especially their biological behavior. So in my work, I used a mesoscopic simulation model, known as lattice postman model, to simulate the fluid dynamics of emulsions. And the lattice postman model is equipped to tackle multiple fluid dynamics equations and conservation laws, including the continuity momentum equations. As you make the system more complex by varying the emulsion droplet geometry or using different complex templates, it is essential to look into using models like the computational fluid dynamics models like the lattice person model, which we talked about. In my current research, I have validated my lattice person model and have looked into some simple results on droplet deformation as we vary across capillary number or shear rate. These results will equip me with my future research on looking into more complex emission droplets. Thank you. Hello. My name is Zach Davis from the Bioinformatics Data Science Program here at the University of Delaware, and I'm working in Dr. Esther Biswas' Fist Lab. Did you know that there are over 2 million people with inherited retinal dystrophies? Many of these diseases are linked to a gene called ABCA4, which has over 4,400 genetic variants. Almost 1,800 of these variants are known as variants of uncertain significance, or VUS. This genetic uncertainty limits patient access to precision medicine -based therapies, entry into clinical trials, and prohibits variants to be used to predict disease prospectively. My study examines 65 of these VUS within the regulatory domains of ABCA4 by using in silica pathogenicity predictors as well as structural analysis to assess their potential impact. By integrating computational tools, I aim to improve variant classification and identify structural patterns that may contribute to disease mechanisms. If you're interested in learning more, come visit me at poster number five. Hello, I'm Neha Sintu, PhD in bioinformatics. I work with Dr. Shove. ADAM9 is the protein that I study. It's interesting because it's upregulated in many diseases, especially almost all the solid tumors. And ADAM9 is also a cell surface protease, which means it lives on the surface of the cell and it cuts other proteins. So what I try to study is to find the substrates, the proteins which are cleaved by and ADAM9. And for this, I do a multi-omics method where I do the RNA -seq, the entire RNA profile of the cell, proteomics, the entire protein profile of the cell, before and after knocking down of ADAM9 in colon cancer cells. And we integrate these with a multi-omics integration method known as O2PLS, orthogonal partial least squares. And with the differences in When the protein changes, we find the substrates and targets of Adam9 and find the pathways which are downstream of Adam9. For more, visit me at poster number six. Thank you. Good morning, everyone. My name is Nikhil Danankam. I am a second-year PhD student under Professor Ulf Schiller. Recent advances in artificial intelligence have revolutionized our ability to analyze, generate, and reconstruct the microstructural images, which is crucial for understanding the PSP, which is the process structure property relationship of the microstructure. Right now I'm working on various GANs architectures and diffusion model architectures to work on micro 2D data set with the end goal of cultivating it to the experimental data sets. Our future goals would be to work on conditional diffusion models using latent diffusion models and include physics-based neural networks to the model. Thank you. I'm Manju. I'm a PhD student in the bioinformatics data science. I work with Dr. Kathy Wu. Today I'm presenting our model case, Finder 2.0, which predicts kinases of human phosphocytes. Kinases are enzymes which are abnormally regulated in a number of human diseases, especially in tumors, and we are predicting the phosphocytes they target. Most existing tools, they look for sequences around the phosphocytes, but in addition we are also looking at the substrate protein's characteristics using knowledge graph embedding and deep learning method. We benchmark our model with other state-of-the-art kinase substrate prediction tools, and we also benchmark the embeddings by comparing with popular protein language models like Prat-T5, ESN2, and ESN3. If you are interested, and if you would like to talk more, please stop by my poster, poster number eight. Thank you. Hello, everyone. So I'm Sid Chaini. I'm a third-year PhD student at the Department of Physics and Astronomy at the University of Delaware, and I mainly work on astrophysics and the kind of objects I'm interested in are the death of stars called supernovae. So supernovae essentially are stars when they die, they result in a cataslemic explosion. And in this explosion, we see that there's a lot of light emitted. The best way to study this is to look at the spectra, which tells us how the brightness changes with the wavelength. Nowadays, we have large-scale surveys where we can collect a lot of spectra. And I am working on a project where I'm trying to do unsupervised learning on this data set of supernovae spectra to see if we can find hidden patterns and reorganize the classifications of different supernovae. If you want to learn more, please stop by my poster, number nine. Thank you. Hello, everyone. My name is Nal Kachow. My presentation focused on mapping of global irrigated area at a final resolution of 30 meter using a combined approach of martial learning and remote sensing. So as you know, irrigation disproportionately contributes to more than 40% of the global food production, while it lies less than a quarter of global level land, and it has huge implication for sustainable land and water management. However, the current existing irrigation datasets are at a coarser spatial resolution, make it very challenging to use it to monitor a field-scale analysis and monitoring. It's considering that we propose to develop a high-resolution 30-meter-gated area mapping globally using a combined approach of machine learning and remote sensing. Since the start of the century, from the year 2000 to 2023, our approach involved a collection of intensive ground truth points, combine it with remote sensing landsat image, curate those ground truth points based on vegetation index and other hydroclimatorical variables and classify each centimeter pixel, whether gated or not, using random forest and other martial learning algorithms. Our preliminary results showed a promising accuracy level as it violated using its test data set and other existing data set. If you want to learn more about it, please stop by poster. Hi, everyone. My name is Nii. I'm a PhD student from the Department of Civil Engineering here in the University of Delaware, and I'll be talking about my work with hurricane wind loss modeling. We understand hurricanes are one of the most devastating natural disasters, which caused a lot of loss to property, and it's very essential for us to be able to accurately predict hurricane losses. However, over the years, due to the lack of insurance claims data and even the credibility and proprietary nature of the data, it's been difficult for us to develop models that predict losses, which has led to the over-reliance on simulation-based models. Thankfully, due to our partnership with the North Carolina Insurance Underwriting Association, we're able to get a hold of insurance claims data, which we've used to develop a two-step model by leveraging machine learning to not only predict loss amount, but first we want to predict the probability of a building incurring loss in the first place, which is not talked about in the community of natural hazards. And using the output of both models, we can estimate the expected loss for each building within our study area so stop feel free to stop by my poster poster 11 to learn more about my work thank you morning my name is Onikachi Williams I'm a PhD student from Delaware State University so my topic I'll be talking about is many using different machine learning method in order to assess the relationship that exists between total suspended solids and other water quality parameters. In Delaware, you see that shellfish farming is actually booming, and we have oyster, we have the blue crab, but very importantly water quality parameters are very, very important, not just only to help the farmers to make informed decisions, they are also very, very necessary for sustainable growth. So what we do, So total suspended solid, which is one of the water quality parameters, we have turbidity, we have temperature, and we have salinity and many more. But it's very, very difficult to be able to estimate or get the values of total suspended solid right in the field. So I started using the standard method. So we tried to compare different machine learning algorithms. We used the neural networks. we did the support vector machine we did the Gaussian processor and from our findings we are able to see that a Gaussian processor regression model was able to produce more and better results so if you want to get more insight in my topic you can stop by at postal number 12 thank you very much i'm a phc student in university of delaware my professor is professor lala barmakin my postal number is 13. Let's be honest, we all not are engaged always while we learn. So sometimes a little break or a pop quiz can help us to learn more effectively. So, but is it possible that our learning tool, our machine, can understand that we are getting, losing our focus and we are getting bored? So here comes the brain signal and the data science. So what we did in our work that we collected oxygen flow during educational game playing. We use functional near-infrared spectroscopy, and we proposed a hybrid deep learning model with CNN and GRU. We predicted the performance score while playing the educational game. And based on the brain signal, that is oxygen flow to our brain, and the performance score, we actually measure two metrics of performance, cognitive report, relative neural efficiency and involvement. So our proposed model shows 73.08 percent accuracy. With this moderate accuracy, we actually find that the actual trend of our relative neural efficiency and involvement follows the predicted one. Maybe I told the opposite one. Predicted actually follows the actual one. so I'm not going to say that if you are interested in my work because I know you are curious about curious about my work so please visit poster number 13 thank you everyone my name is Logan Halley I'm a PhD candidate in bioinformatics data science and I'm also the CSO and founder of Sinthera and no matter what hat I'm wearing my work is always focused on proteins proteins are of course these fantastical biological molecules that are pretty much the active components of biology they regulate almost everything through their physiochemical interactions And despite this, less than 1% of 1% of the positive protein sequences are labeled or annotated. And so what's on my poster today is a new kind of transformable readable language called the annotation vocabulary, which is a better way to kind of leverage these sparse labels so that we can better label and annotate proteins in deposited data sets. And so we showcase some models leveraging this with state-of-the-art annotation results. And we also have a generative model which showcases the first natural -like, de novo-generated protein sequences that are far outside the training set distribution. Thank you very much. Good morning, everyone. I'm Nikhil, also in Dr. Schiller's lab. And today I'll be giving a talk, or a poster, about my investigations into stimuli -responsive droplets. The essential question we're seeking to answer is how can we use stimuli-responsive droplets as a delivery mechanism for pharmaceuticals? So what we've observed with these droplets is that if you stabilize them with ellipsoidal particles, which are non -spherical particles, and in our case, rod -like particles, what we see is that as you increase the magnetic field, we have a greater shape structural response that is characterized as a flattening of the droplet. And we also show why this is the case in the poster, and we also notice that we see some accumulation of particles in the equatorial region of the droplet, but not in the tips. So what we estimate to see is if we have multiple droplets together, we might see some droplet coalescence. So if we're interested to learn more about how we leverage computational fluid dynamics to measure droplet deformation in these systems, do visit my poster at Poster 15. Thank you. Good morning, everyone. I'm Rajas Mahindari. I am a PhD student in the chemical engineering department. I work with Professor Dion Lacos. As we all know, the world is facing a big problem of handling plastic waste. Over 400 million tons of waste is produced every year. And one of the most important ways of getting rid of this waste is by chemically recycling it. In our work, we use a combination of density functional theory and high-performance computing using molecular dynamics with parallel tempering or replication to model these polymers on catalyst surfaces which goes a long way into designing more rational catalysts for plastics recycling please visit my poster if you are interested in learning more about this thank you morning everyone i am a phd candidate at the department of chemistry and biochemistry here at ud and today at my poster i'm going to present a framework that we developed to build comprehensive chemical maps of surfaces based on hyperspectral imaging. So the novelty of this methodology is that it not only maps the chemicals in the surface but also estimates the number of these unique regions. So today I'm of course going to present three different cases that I applied this framework. The first one is the mapping of archaeological painting dated from the 2nd century. I'm also presenting the Rames microscopy of meteorites impacts. And the third one is the photoluminescence mapping of quantum materials. So thank you very much. My name is Michelle Eshelman. I'm a PhD candidate in the entomology and wildlife ecology program in Dr. Jeff Buehler's lab. So I'm a bit of an odd duck here being from that program. I am going to be talking to you today about the tracking of wildlife across the year. So if you're a bird watcher, you might have noticed bird populations are declining. We've lost about 30% of our birds in the last 50 years. And we know that during migration, it's a time of high mortality for songbirds. So we want to study that period of their life to understand maybe where these population declines are occurring. I study eastern toies, a species that is steeply declining and relies on early successional forests for its survival and breeding. We track these birds using small radio transmitting tags. and our data set generated over 900,000 detections of our birds we found that they are using overwinter the Delmarva Peninsula for our East Coast breeding populations and I'd love to talk to you more about our study at poster 19 thanks hello everyone I am associate professor from Turkey I am a visiting scholar as a postdoctoral researcher at Disaster Research Center and I share my research on disaster preparedness and disaster education. I conducted a bibliometric analysis of 224 academic papers from Scopus Center of Science. My findings show a significant rise in disaster related publication after 2018, particularly due to the COVID-19 pandemic. My study identified the most effective works in the field, highlighting key gaps in psychological resilience and institutional readiness. I invite you to visit my poster to explore the data, visuals, citation trends, and potential pathways for future interdisciplinary collaborations. Thank you so much. I'm Michael Liebman. I also direct a nonprofit focused in women's health. To change medicine, we have to move from correlation to causality to really be able to have an effect. That means what we've done is try to evolve from purely data-driven modeling to model-driven data analytics. That enables us to move from the world of known unknowns to start to approach unknown unknowns. The biggest unknown unknown that we're dealing with is in women's health, and that's the natural transition of menopause. We have a normal situation that affects 51% of the world's population directly, 49% of the world's population indirectly, and so everyone has an impact of menopause, and yet we have very poor data, very poor understanding, and don't acknowledge that women go through this individually, almost independently, and your hot flashes are not the same as your neighbor's or your friend's hot flashes. And that also means that at the end of the day or the end of the transition, your risk for breast cancer, for cardiovascular disease, and osteoporosis are your risks and not the same as your friend's and neighbor's, and that's what we try to build models for. Thank you. Great. Let's thank our Lightning Talk presenters one more time. And now we'll begin the poster session itself. And so please feel free to visit the posters that we heard about and be sure to vote afterwards. Once again, the voting forms are at the check -in desk. And we'll accept votes, I believe, up until 1.15 today. And after that, for our poster presenters, be sure to stick around until the end of the day when we'll be announcing the best poster awards. Thank you.
Lightning Talks
From Jason Felton April 11, 2025
0 plays
0
0 comments
0
You unliked the media.
Welcome to the Delaware Data Science & DARWIN Computing Symposium, hosted by the University of Delaware Data Science Institute. This video was recorded on April 7, 2025, at the Tower at STAR Audion (UD Star Campus).
The 2025 Data Science & DARWIN Computing Symposium features keynote talks, a panel on “Advancing Data Science and AI for Grand Challenges”, poster presentations, research computing tutorials, and networking opportunities.
The symposium also features presentations of results from the DSI Competition “Mapping DSI Faculty Networks and Expertise at the University of Delaware” and an awards ceremony with competition and poster awards.
- Tags
- Date Established
- April 07, 2025
- Appears In
Link to Media Page
Loading
Add a comment