Illinois Mathematics and Science Academy
Illinois Mathematics and Science Academy: DigitalCommons@IMSANot a member yet
9795 research outputs found
Sort by
International Conference on Machine Learning (ICML) 2025 in Vancouver
No full textNeel Shanbhag ’26 and Laksh Patel ’26 presented their AI research at the International Conference on Machine Learning (ICML) 2025 in Vancouver and won the best Poster for Data in Generative Models Award.https://digitalcommons.imsa.edu/homepage/1078/thumbnail.jp
IMSA 2025 Profile
Get PDFThe nationally-ranked Illinois Mathematics and Science Academy (IMSA) develops creative, ethical leaders in science, technology, engineering and mathematics. As a teaching and learning laboratory created by the State of Illinois, IMSA enrolls academically talented Illinois students in its tuition-free residential Academy for grades 10-12. Students are challenged with a rigorous curriculum designed to develop them into problem solvers and critical thinkers. Notable IMSA alumni include YouTube Co-Founder Steve Chen, PayPal Co-Creator Yu Pan, Yelp Co-Founder Russell Simmons, SparkNotes and OkCupid Co-Founder Sam Yagan, and Hearsay Social Founder, Clara Shih
Rigging the Role: Probability, Fairness, and Winning Every Time
Get PDFStep into the world of dice games with a twist! In this session, you’ll explore the fascinating interplay of fairness, strategy, and mathematical reasoning. Through a hands-on dice rolling competition, discover how to analyze theoretical and experimental probabilities and investigate what makes a game “fair.” Then, dive into the ultimate challenge: designing a “rigged” die to outmatch others in a roll-off. Perfect for middle school math classrooms, you\u27ll leave with ideas to captivate learners while connecting probability to the real world. Join us to learn, play, and win—every time
Development of a Baculovirus-based Packaging System for Efficient Recombinant Retrovirus Production
No full textRetroviral vectors are commonly used to generate stable cells to express transgenes. However, packaging high-titer retroviruses is technically challenging due to variations in co-transfecting the packaging cells with multiple plasmids that express genes essential for retrovirus production, leading to inefficient and inconsistent virus production and dramatic virus titer fluctuations. The objective of this study is to investigate whether or not baculovirus (BV)- mediated delivery of gag-pol-env and VSV-G would produce high-titer retroviruses. If successful, such a system would significantly simplify retrovirus-making, which is crucial for effective stable transgene expression in biomedical research. Experimentally, the gag-pol-env and VSV-G expression cassettes were first cloned into BV transfer vectors, and the resultant constructs were verified by DNA sequencing. The BV transfer vectors were transformed into DH10Bac bacterial cells, which harbor the BV genome that expresses Tn7 transposases. After the Tn7 transposition reaction, positive (white) clones were selected from X-gal/IPTG blue- white agar plates. The BV vectors for gag-pol-env and VSV-G were obtained. After verification, these BV vectors were transfected into Sf9 insect cells to produce BV viral particles. We have successfully obtained BV-gag-pol-env and BV-VSV-G baculoviruses, and are working on amplification to create high titer BV viruses for function tests and retrovirus production
Comparing the RSP Accuracy of Dual-Energy to Conventional Single-Energy CT for Potential Reduction of Clinical Safety Margins in Proton Therapy
No full textIn proton therapy, treatment planning is reliant on the accurate prediction of relative stopping power (RSP) to calculate the proton range necessary to sufficiently treat the target volume and minimize the dose given to healthy tissue and organs at risk. However, uncertainties resulting from approximations made in the calculation of RSP from CT numbers necessitates the use of 2.5%-3.5% range uncertainty margin. This study aimed to evaluate the potential reduction of this uncertainty margin using Dual-Energy CT (DECT) by comparing the RSP accuracy of an ultra-fast switching KV DECT to conventional Single- Energy CT (SECT). Reference RSP values of tissue-mimicking cylindrical plugs and fresh, postmortem lamb tissue were obtained using MLIC measurements and proton CT respectively. RSP values were then derived from DECT and SECT of both the tissue-mimicking plugs and the lamb tissue and their percentage difference from the reference RSP values was calculated. Overall, DECT demonstrated a lower mean absolute percentage error than SECT for all tissue types in both tissue mimicking phantoms and in animal tissue, potentially implying that the use of DECT could reduce uncertainty in treatment planning
Biomechanical Mechanisms of Tongue Movement During Mastication
No full textMastication, or mammalian chewing, is the cyclic breakdown of the food bolus between the molars. Effective chewing consists of tight coordination of lips, cheeks, jaw, and tongue modulated via sensorimotor integration of tongue movement and sensation. Compromised chewing leads to risks, including digestive issues, malnutrition, and potential impacts on mental well-being and cognitive function. However, the biomechanical mechanisms driving tongue movement during mastication, remains poorly understood. In this study, we evaluated the contributing muscles toward tongue movements during mastication using biplanar videoradiography following the XROMM workflow. Using CT scans to visualize the hyolingual anatomy of opossums, we discovered that the tongue exhibits complex movements, including flexion and rolling, which are essential for effective food manipulation and bolus formation. In vivo observation of quantified marker movement reveal that the genioglossus, hyoglossus, and styloglossus muscles play crucial roles in tongue retraction and depression, while the superior and inferior longitudinal muscles facilitate fine-tuned shaping and elevation of the tongue during chewing cycles. Our findings may facilitate new treatment exercises during rehabilitation for patients with neuromuscular disorders affecting oral motor function
AI-Driven QAM Transceivers: Enhancing Wireless Communication with Machine Learning
No full textQuadrature Amplitude Modulation (QAM) is widely used in modern wireless communication systems to transmit data efficiently. Conventional QAM transceivers rely on specialized hardware to swiftly and accurately modulate, transmit, and demodulate signals. However, hardware-based transceivers are costly and slow to adapt to evolving technologies. In this research, we examine the potential of replacing hardware with AI-based models in QAM transceivers while maintaining performance standards. Using a 16QAM graycode constellation, we simulate a QAM transceiver and generate a dataset to train machine learning models for signal decoding. We assess the model’s accuracy, operational speed, and functionality against traditional hardware receivers in real-time processing. The outcomes of this study support costeffective, adaptable, and intelligent communication systems that optimize spectrum efficiency and enhance the reliability of modern wireless networks
A Study of Intrinsic Excitability for the Hodgkin-Huxley Neuron Model
No full textComputational models have commonly been used to study neuropathologies, allowing for a realistic interpretation of behavior. This study investigates the single-cell Hodgkin-Huxley model to develop a richer understanding of the neuron’s resting state during cognitive activity. Using the Simulator for Neural Networks and Action Potentials (SNNAP) computer program, sodium and potassium conductance are modulated from a quiescent state until observed excitability, becoming an underexplored study in micro-instabilities. The model neuron was quiescent under normal membrane conductances (gNa = 120 mS/cm², gK = 36 mS/cm²) with no external current injection. However, a conductance change as small as 0.001 affected the excitability of the neuron. Therefore, given the goal to locate these conductance thresholds, this study looked at various SNNAP parameters, including the membrane voltage (V), its time derivative (dV/dt), and the voltage dependence of ionic currents (Ivd) for both Na and K ionic channels. This study found significant differences in excitability after the incremental change in sodium conductance and the comparable decrement in potassium conductance. Altogether, these findings show how small modifications in conductance parameters can lead to action potentials without current injection, informing researchers of the realistic thresholds that larger neural networks should maintain in computational neuroscience
Efficacy of Adversarial Attacks on Traffic Sign Recognition Models Presenter
No full textAdversarial learning is a critical area of research that examines the vulnerabilities of machine learning models to carefully crafted attacks. In safety-critical applications such autonomous driving, adversarial attacks on traffic sign recognition systems pose significant risks, potentially leading to severe consequences such as crashes.
This study explores various adversarial attack strategies, including white-box and black- box methods, to assess their impact on the traffic sign classifiers used in autonomous vehicles. Techniques such as the Fast Gradient Sign Method (FGSM), Projected Gradient Descent (PGD), DeepFool, Square Attack, and more are analyzed for their effectiveness in misleading recognition models. Additionally, we evaluate state-of-the-art defense mechanisms, including adversarial training, to enhance model resiliences. By benchmarking attack efficiency and mitigation strategies, we aim to contribute to the development of safe, reliable, and secure autonomous driving systems