University of Illinois Urbana-Champaign
Illinois Digital Environment for Access to Learning and Scholarship RepositoryNot a member yet
123813 research outputs found
Sort by
Monotone Dynamics in Network Control
Get PDFMonotone dynamical systems, which preserve a partial ordering on the states, have a long history; their theory may provide a path to analyzing network control systems, distinct from the more standard methods of optimization or Lyapunov-based control. In this note, after a brief review of some of the main concepts in monotone systems, we survey a series of problems in network control which are amenable to this approach. Using ordinary differential equations, the method applies to dynamic consensus, cooperative load balancing and scheduling of a fixed set of queues, and certain capacity-scaled models of peer-to-peer file sharing. Other applications which require monotonicity in infinite dimensions, are indicated as partially open in this approach: load balancing in the regime of a large number of queues, and transport PDE models for queues under general file-size distributions
The Pulse-Shaping Trojan: Refinements and New Results
Get PDFRecently, an insidious wireless hardware vulnerability was revealed that surreptitiously exfiltrates information from a wireless node by manipulating a transmitter’s pulse-shaping filter. This paper introduces refinements and important new results for the pulse-shaping Trojan that further clarify its threat level and its clandestine nature. Error Vector Magnitude (EVM) is introduced as the metric for the distortion induced by this Trojan, replacing the inner product as a measure of the stealth of the rogue signaling. A Viterbi successive interference cancellation scheme is studied for the rogue receiver, to recover the rogue payload in the presence of the higher-power legitimate transmission. The differences and relative merits of the Viterbi receiver with respect to LMMSE interference cancellation is discussed in the context of Trojan signaling
Estimating Error in Natural Distribution Estimation
Get PDFGiven i.i.d. samples from an unknown discrete distribution, the goal of distribution estimation is to construct an accurate estimate of the underlying distribution. Natural distribution estimators assign one probability estimate to all letters occurring with the same frequency, and this is well-justified for i.i.d. models. However, natural estimators can be significantly erroneous for low frequency or missing (frequency 0) letters in large alphabet scenarios. In this work, we introduce a statistic that captures the unavoidable error at a particular frequency of any natural distribution estimator. For this proposed error statistic, which depends on the distribution and the samples, we provide an estimator that is non-linear in the prevalences (frequencies of frequencies). We show that the proposed estimator has low bias and is consistent, and can be used to ascertain if the distribution restricted to letters of the same frequency is close to uniform. Our approach is validated through simulations on synthetic and natural language data
A Modular Federated Suite for Low-Rank, Expressive, and Efficient LLM Fine-Tuning
Get PDFThis works presents a modular federated suite of four methods—LoRA-SilverBullet (LoRA-SB), ABBA, Fed-SilverBullet, and FedEx-LoRA that together enable low-rank, expressive, and resource-efficient fine-tuning of large language models (LLMs). LoRA-SB approximates full fine-tuning within low-rank subspaces via a principled initialization strategy, provably preserving gradient directions and reducing trainable parameters by 27–90× without any additional hyperparameter tuning. ABBA reparameterizes weight updates as the Hadamard product of two low-rank matrices, and formally increasing expressivity under a fixed parameter budget. FedEx-LoRA introduces a lightweight residual correction to recover exact LoRA adapter updates under standard federated averaging, preserving efficiency with minimal overhead. Fed-SB leverages the LoRA-SB’s low-rank update with FedEx-LoRA’s exact aggregation in differentially private federated learning, and cuts communication costs by up to 230x. We will detail theoretical results on convergence, reconstruction bounds, communication complexity, and privacy loss, alongside empirical evaluations on reasoning and language benchmarks. This suite offers a principled path to deploy LLM fine-tuning in resource-constrained and privacy-sensitive federated environments
On Explicit Families of Outer Bounds for Broadcast Networks
Get PDFThe cut-set bounds arise from the graph-theoretic notion of a cut and are used as outer bounds for broadcast networks. Although these bounds are tight for broadcasting a single message, they are loose when multiple messages are broadcasted. The generalized cut-set bounds (GCBs) were proposed in [1] in implicit form as a broader class of outer bounds. In this work, we build on the GCB framework and present two families of explicit outer bounds with associated structure. We also present two techniques to modify GCBs that we call GCB-mappings. These are mappings that the class of GCBs are closed under. They enable customizing known outer bounds to generate novel bounds with similar structure. We present a systematic approach for proving outer bounds for broadcast networks using these results. In addition to the concise proofs it yields, this approach also reveals common structure shared across the inequalities within a capacity region. We focus on the combination network (CN) as a special case to demonstrate our approach. First, we revisit the well-known capacity region for the 3-user complete message set and present an alternative converse proof using a single explicit family and a GCB-mapping. Next, we turn to a 4-user generalization of this problem that we call the cubic message set, which includes an additional receiver that decodes all seven original messages, as well as an additional unicast message. We show that our outer bounds are tight when specialized to this setting and prove the converse of the capacity region. Finally, we present an explicit outer bound for the 5-user extension of the cubic message set
GLP-1 Receptor Agonists: A Potential Treatment for AUD
No full textAlcohol use disorder (AUD) is a chronic substance use disorder characterized by uncontrolled alcohol consumption. According to the National Institute on Alcohol Abuse and Alcoholism, it affects roughly 28.9 million people in the US (2024) and was linked to 2.6 million preventable deaths worldwide in 2019 (World Health Organization, 2024). One major roadblock in treating AUD is the limited availability of prescribable medications-only three are currently approved in the United States. Consequently, a major focus of current research is finding suitable therapeutic agents. Recent clinical studies suggest GLP-1 receptor agonists (GLP-1 RAs), which are normally used to treat Type 2 diabetes, may be potential candidates
Adaptive Plasticity of the Colorblind Brain: A Model for Sensory Compensation
No full textColor vision deficiency (CVD) or color blindness results from X-linked recessive genetic mutation that decreases or impairs the expression of cone cell photoreceptors essential for normal color perception. As a result, individuals with color blindness are unable to distinguish certain colors or hues in the same way as individuals with typical color vision. On a molecular level, the most common forms of CVD arise from the absence or malfunction of one type of cone cell in the retina, which reduces sensitivity to specific wavelengths of light. This disruption in normal color processing leads to altered color perception, often making daily visual tasks more challenging. However, the colorblind brain can adapt to these perceptual differences through neural plasticity. Recent neuroscience research indicates that visual cortical areas V2 and V3 are particularly involved in cortical reorganization in individuals with CVD. Additionally, at the cellular level, structures such as rods, intrinsically photosensitive retinal ganglion cells (ipRGCs), and neurons in the lateral geniculate nucleus (LGN) may contribute to compensatory neuroplastic responses to altered visual input. By using current research on the adaptive plasticity of the brain in color blind people, scientists can further the potential of neural training for rehabilitation and therapeutic strategies targeted to treat brain trauma, injuries, or other visual impairments
Fear on Repeat: Examining the Impact of PTSD on the Amygdala
No full textPost-traumatic stress disorder (PTSD) is a psychiatric disorder inflicted by experiencing or witnessing a traumatic event, triggering a variety of symptoms. Individuals with PTSD can experience symptoms such as nightmares, flashbacks, and detached behavior, leading to the inability to function normally, particularly in social or family life (Iribarren et al., 2005). These symptoms can affect a person for a lifetime, emphasizing the importance of this disorder
How would AI define neighborhood boundaries? A comparison with human-crowdsourced data
No full textGenerative artificial intelligence (AI) has started to be considered a cost-efficient alternative for geospatial and urban surveys, but there remains a critical need to evaluate how closely AI-generated outputs align with human responses. This paper compares responses from ChatGPT and residents in defining neighborhood boundaries, a long-standing challenge in urban studies with no single correct answer and typically relies on input from resident surveys. Our analysis focuses on both the defined boundaries and areas that are rarely covered by any boundaries. Our results show that ChatGPT tends to generate neighborhood boundaries with less variability in extent and geographic coverage compared to crowdsourced boundaries, potentially favoring more standardized representations. Additionally, we find that AI-generated boundaries are less likely than human efforts to cover areas with lower population density and higher percentages of non-White and Hispanic populations, reflecting potential biases. These findings highlight the need to critically evaluate generative AI’s potential to supplement human respondents in urban and spatial applications while carefully considering its limitations, particularly regarding bias and representation
The Consequential Effects of Sleep Quality on the Academic Performance of University Students
No full textSleep is a vital biological process essential for proper physiological and cognitive functioning. While its true purpose remains largely theoretical, sleep deprivation has been proven to impair numerous brain functions, particularly in college students who are increasingly susceptible to irregular sleep patterns. Drawing on evolutionary and neurological theories, such as the restorative, synaptic homeostasis, and brain plasticity theories, the article examines the impact of sleep deprivation on cognitive processes, memory retention, attention span, and overall brain connectivity. Through recent studies utilizing tools such as MRI and attention network tests, a consistent decline in brain activity and memory function was observed in sleep-deprived individuals. These effects are especially prevalent in university settings, where academic pressures, lifestyle changes, and increased substance intake contribute to deteriorating sleep quality. The findings highlight that inadequate sleep not only diminishes students’ ability to retain and process information but also places them at higher risk for academic failure and long-term health consequences. Ultimately, the article emphasizes that sleep is not only necessary for survival but also fundamental to academic success and cognitive resilience