UC Berkeley

Quantum locally recoverable codes (QLRCs) have recently gained attention as a framework for achieving efficient quantum storage with local recovery capabilities. Analogous to their classical counterparts, QLRCs allow a lost qudit to be reconstructed using only a small subset of other qudits, thereby reducing the resource and operational overhead in recovery. In this work, we extend the study of QLRCs by considering (r, δ) QLRCs characterized by locality parameter r and local distance δ ≥ 2. We present constructions of both random and explicit (r, δ) QLRCs, including explicit families based on the quantum Tamo—Barg construction. We also present an efficient decoding algorithm for these quantum Tamo–Barg codes.Furthermore, we introduce quantum hierarchical locally recoverable codes (QHLRCs), which extend local recovery to multiple hierarchical levels. For any integer h ≥ 2, we construct both random and explicit h-level QHLRCs—the latter being h-level quantum Tamo—Barg codes—and establish a Singleton-like bound for these codes using a CSS framework built from dual-containing classical codes. These results advance the theoretical foundations of quantum erasure recovery and contribute to the design of efficient quantum storage architectures.

One of the largest obstacles in modern drug discovery is that a significant portion (>90%) of the proteome is considered “undruggable,” in that these proteins lack a characterized, functional binding pocket or ligandable hotspot which small molecules can bind to and modulate the protein’s function for therapeutic benefit. To overcome such disease-causing proteins, targeted protein degradation (TPD) strategies have arisen, where the cell’s endogenous degradation machinery is hijacked to ubiquitinate and degrade the classically undruggable protein. Molecular glue degraders serve as a promising modality to achieve TPD. These are monovalent compounds that induce the proximity of a target protein with a component of the ubiquitin proteasome system to degrade the protein of interest. While our research group has previously identified a fumarate-based electrophilic handle that covalently modifies the E3 ligase RNF126 to enable degradation of multiple protein targets, the high intrinsic reactivity and cytotoxicity of the fumarate handle limited its translational utility. This work describes the development of an optimized and metabolically stabilized RNF126-targeting covalent handle incorporating a trans-cyclobutane linker that exhibits reduced glutathione reactivity and diminished cytotoxicity while retaining robust degradative activity. Using BRD4 as a benchmark target, we demonstrate that this optimized handle yields a potent and selective BRD4 degrader whose activity is dependent on RNF126. We further extend this strategy to the androgen receptor (AR) and its clinically intractable splice variant AR-V7, demonstrating selective degradation of both AR and AR-V7 in androgen-independent prostate cancer cells and robust inhibition of AR transcriptional activity that surpasses the established AR antagonist enzalutamide. Together, this dissertation establishes a generalizable, chemistry-centric framework for converting small-molecule ligands into covalent molecular glue degraders, offering a roadmap for exploiting event-driven pharmacology against the most intractable targets in the human proteome.

The works investigate reliability and performance enhancement for ultra-scaled advanced CMOS and memory devices through experimental characterization, physical modeling, and device simulation. The study first examines random telegraph noise (RTN) in advanced transistors under cryogenic operation, revealing the strong dependence of carrier trapping dynamics on temperature and bias conditions. A physics-based RTN modeling framework, combined with experimental results, is developed to predict RTN behavior in nanoscale transistors. In addition, the impact of ultra-thin aluminum incorporation within high-k metal gate stacks (HfO₂/TiN) is investigated. Electrical characterization and material analyses reveal that aluminum modifies the effective work function through oxygen scavenging from the TiON layer. The reliability trade-offs associated with Al incorporation in the metal gate are further clarified. Next, to mitigate the side effects of Al-incorporated gates, an oxygen insertion (OI) technology is proposed as an alternative approach for controlling the flat-band voltage. Lastly, a silicon-germanium/silicon (SiGe/Si) heterojunction drain transistor is proposed for 3D NAND flash memory to enhance gate-induced drain leakage (GIDL) and improve erase speed through increased band-to-band tunneling. Simulation results demonstrate significant improvement in erase performance without degrading inhibit-mode operation.

Artificial intelligence (AI) is increasingly used by individuals, firms, and organizations around the world, offering an opportunity to expand capabilities and improve livelihoods. Yet adopting and adapting to any new technology is difficult, and these challenges are compounded when the technology is as complex as AI. This dissertation examines how AI could shape strategic decision processes, with a focus on understanding the implications of AI system capabilities for low-resource contexts. Underlying the dissertation is the idea that if we want AI systems to have a positive impact, we must begin by understanding model performance. My first two essays explore this within the context of AI-generated advice: if we can understand where generative AI provides high-quality advice, then we can identify cases where AI should be deployed and cases where additional input may be needed in order to be beneficial to the user. I start by exploring this within the context of a field study in Kenya which varies the extent to which individuals receive advice from an AI system as opposed to traditional human advice. Results from this study suggest that AI-generated advice and expertise may reinforce each other, while highlighting the risks that can emerge if individuals begin to use AI at the expense of vetting the advice they receive by engaging with external sources of information. I then explore the extent to which the recommendations produced by AI systems may inadvertently privilege some contexts over others. Specifically, I test whether the default response from AI systems provides higher quality advice to individuals living in high- versus low-resource contexts. Results from this study provide evidence that default responses from AI are more aligned with high-resource contexts, which I document both within and across countries. However, I find that much of the difference in the quality of responses provided by AI stems not from fundamental differences in the model's underlying capabilities, but instead from the default context assumed by the model. Finally, I examine the performance of AI in the context of a specific use case, evaluation. I develop a test of whether AI systems prefer proposals that were written by AI through a novel design which holds proposal quality constant, varying only the extent to which language inordinately used by AI models is included in the proposal. I provide evidence of self-preference in model evaluation, and show that these distortions are substantial enough to change the ranking of proposals, while also highlighting that because of the low cost of using AI for evaluation, it can still be beneficial to use AI systems, even given these errors, if the alternative evaluation strategies are sufficiently costly. Together, these essays highlight the potential of generative AI systems to improve people's livelihoods, while also cautioning that understanding the capabilities of these systems is fundamental to using them effectively. By better understanding AI system capabilities, we can identify when more contextual information or guidance is needed, and when AI advice, while imperfect, may still be efficient due to its lower cost. As the capabilities of AI systems continue to rapidly improve, ensuring that these tools are deployed with the right enabling resources will be an important part of realizing their potential for people around the world.

This dissertation studies Large Language Model (LLM) inference through the lens of statistical decision-making theory, with the goal of advancing prevailing inference paradigms along two principal axes: scalability and provenance. In the domain of scalability, this dissertation investigates the efficiency of inference-time scaling paradigms and parallel decoding schemes. We formally characterize the sample complexity of self-consistency and best-of-n sampling methodologies, and demonstrate how self-correction can substantially expand the expressive power of Transformer architectures in multi-task settings. We analyze the information-theoretic bottlenecks associated with parallel sampling in diffusion language models, and introduce Explore-Then-Exploit, a scalable decoding strategy that improves inference throughput while preserving generation quality. In the domain of provenance, the dissertation develops theoretical foundations and practical schemes for statistical watermarking. We mathematically formalize statistical watermarking as a hypothesis-testing problem, providing a comprehensive characterization of the optimal Type II error, token efficiency, and robustness against perturbations. Guided by the theoretical insights, we propose SEAL, a semantic-aware watermarking scheme demonstrating strong detection efficiency and tamper resistance. We further generalize the statistical watermarking framework to support anytime-valid detection and derive the optimal e-value-based detection scheme. Collectively, our work connects theoretical insights with algorithmic innovations for understanding and improving the efficiency and responsible use of LLM inference.

The Death of the Asylum: How Federal Policy, Popular Culture, and Drugs Killed the American Mental Hospital (1940-1946) is a historical study of three interwoven elements that led to the end of asylums or psychiatric institutions in mid-20th century America. The dissertation is an interdisciplinary project that brings together methodologies from the history of architecture, psychiatry, and law to tell the story of the extinction of an American institution and building type. The work argues that 1946 was a pivotal year for the beginning of “deinstitutionalization,” defined as the phasing out of institutional treatment of mental illness in asylums established in the late 19th-century and towards an outpatient medical model. This shift fundamentally altered the profession and practice of psychiatry.The first chapter explores the legacy of journalistic exposés and literature starting in the late-19th century by writers who were hospitalized for feigned or real mental illness. These writers, such as Nellie Bly and Mary Jane Ward, used their experience in the asylum to write books and articles that chronicled their often-shocking journeys and critiqued the treatment they received. The chapter begins with a historical overview of the original ideals and humanitarian intentions of the 19th-century asylum. A new architectural typology in the late 1800s, this building type was rooted in the healing power of nature and the built environment. This historical and architectural exposition provides a stark contrast to what became of asylums by the mid-20th century, which the conscientious objectors of Chapter 2 sought to change. The chapter then primarily analyzes Ward’s popular novel The Snake Pit (1946) and its Hollywood film adaptation (1948), and Ward’s courageous but costly involvement in mental health reform. The second chapter begins in 1940 with the Selective Training and Service Act to tell the story of the four conscientious objectors who created the Mental Hygiene Program of the Civilian Public Service during World War II. I argue that these leaders—Leonard Edelstein, Hal Barton, Phil Steer, and Will Hetzel—changed the trajectory of mental health treatment in the United States through their activism. After being assigned to work in understaffed asylums as part of their alternative service because of their pacifist beliefs that protested the war, these conscientious objectors led a movement to document widespread abuses among hundreds of mental hospitals, analyze existing mental health commitment laws, and educate and retrain future employees of mental hospitals.The final chapter follows the career of journalist and wartime correspondent Albert Q. Maisel, who worked with the conscientious objectors in Chapter 2 who had formed the National Mental Health Foundation at the conclusion of World War II. In partnership with the conscientious objectors, Maisel procured eye-witness testimonies and haunting photographs to write a scathing and damning exposé article titled “Bedlam 1946,” which was published in Life Magazine on May 6, 1946. The chapter concludes with the National Mental Health Act (NMHA), which was signed into law by President Harry Truman on July 3, 1946. The Act funded the creation of the National Institute of Mental Illness (NIMH), catalyzed the development of psychopharmacological treatment, and was the final nail in the coffin for the American asylum. Through archival research, this dissertation demonstrates how, in 1946, the dream of the curative mental asylum was over.

Robots hold the promise of assisting humans in unstructured, everyday environments, yet today's robot systems remain far behind humans in sensorimotor control. They are either dexterous but brittle, or generalizable but clumsy. The current dominant approaches in robot learning rely heavily on scaling human demonstrations; this dissertation argues that such approaches are fundamentally limited by the embodiment gap between humans and robots, as well as by the scarcity of high-quality data, especially for dexterous manipulation. Instead, I advocate for a shift toward Embodied Intelligence from Autonomous Experience: a paradigm in which robots acquire increasingly dexterous and generalizable skills through active exploration and self-improvement. This dissertation presents three primary contributions toward making this paradigm practical. First, I introduce methods for learning dexterous priors from human guidance, including a low-latency teleoperation system and a visuotactile policy learning pipeline that help bridge the human–robot data gap. Second, I develop principled techniques for practical sim-to-real reinforcement learning, demonstrating that complex, contact-rich manipulation skills can be learned efficiently without massive compute or photorealistic simulation. Third, I present frameworks that unify and extend existing algorithmic components toward more capable real-world robot systems. Together, these contributions enable robots to perform fine-grained, contact-rich, long-horizon tasks with a level of dexterity and generalization previously difficult to achieve with either model-based or learning-based methods. More importantly, they demonstrate a path toward robot learning systems that not only learn from humans, but also improve through autonomous embodied experience.

Heterogeneous catalysis is a highly important research area in chemistry that can address the climate crisis. Advances in computational chemistry tools, particularly those based on density functional theory (DFT), have been employed to uncover the underlying physics of the operation of catalysts and have contributed to their more efficient design. In this dissertation, we develop and apply computational schemes using density functional theory (DFT) to a broad range of topics relevant to heterogeneous catalysis. In the first half of the dissertation, we combine orbital-optimized (OO) single-reference methods with the one-electron exact two-component (X2C) Hamiltonian to model the ionization of core-electrons. In Chapter 2, we present a non-orthogonal, quasi-degenerate perturbation theory (NO-QDPT) approach for treating the spin-orbit coupling of core-electrons. This scheme is applied to a broad range of ionizations across the periodic table in order to assess its performance and identify its limits. The NO-QDPT approach demonstrates near-quantitative agreement with experimental core-electron binding energies for third-row elements, though its accuracy decreases for later first-row transition metals. In the second half of the dissertation, elementary ion-transfer reactions are modeled via Constant Electrode Potential (CEP) calculations. In Chapter 3, the adiabatic, reversible electrodissolution of anodically polarized silver is modeled. By employing a hybrid implicit-explicit solvation model and the CEP protocol, the study successfully predicts reaction trajectories and free energy barriers that align with temperature-dependent experimental data. The findings reveal that the reaction barrier is a delicate competition between solvation, metal-metal bonding, and image charges. In Chapter 4, the role of hydroxide-mediated intermediates in cathodic copper degradation during the carbon dioxide reduction reaction (CO2RR) is studied. CEP calculations show that hydroxide-mediated dissolution of exposed surface Cu atoms is potentially viable. Specifically, the formation and subsequent dissolution of [Cu(OH)2]- from surface defects is found to be energetically accessible and is facilitated by the formation of an image quadrupole. Together, these findings demonstrate that refined computational treatments of the electrode surface can provide a fundamental understanding of the surface chemistry governing the durability of heterogeneous systems.

Sampling algorithms based on Markov chain Monte Carlo (MCMC) are ubiquitous across theory and practice. The success of MCMC algorithms is typically established by upper bounding their mixing time, and over the past several decades a rich theory for doing so has emerged.However, classical mixing time based arguments fail to explain why MCMC has been so successful for statistical inference and the scientific simulation. The classical notion of rapid worst-case mixing is often too stringent of a notion to apply to these settings. Instead, one needs to analyze warm start mixing, or analyze the inference and optimization performance of MCMC before it has mixed. In this thesis, we build new theory in these directions to explain MCMC’s success in these difficult-to-understand settings.We then pivot to an entirely different subject matter altogether: that of synthetic data in modern machine learning. We study the counterintuitive success of synthetic data in regimes where the data is generated from models that are weaker than their students. We also probe the limitations of synthetic data to elicit certain reasoning behaviors from language models.

Herein, I describe the development of a Pd-catalyzed C–C bond cleavage/vinylation/Mizoroki–Heck cyclization cascade reaction and its application to total syntheses of 14- and 15-hydroxypatchoulol, alongside our synthetic studies toward plumisclerin A.Chapter 1 reviews Pd-catalyzed Mizoroki–Heck reactions in total synthesis. In addition, it highlights recent contributions from the Sarpong group on Pd-catalyzed C–C bond cleavage/C–C bond forming strategies in synthetic methods development and total synthesis of natural products.Chapter 2 details our work on employing gem-dichloroalkene electrophiles as cross- coupling partners in palladium catalysis. Through the use of various phosphine-based ligands and careful substrate design, we achieved high selectivity among four possible products of the envisioned Pd-catalyzed cascade reaction: the vinylation product, bicyclo[2.2.2]octane, bicyclo[3.2.1]octane, and tricyclo[3.2.1.0]octane. Our mechanistic proposals regarding the observed product selectivity are supported by data science and computational studies conducted in collaboration with the Sigman group and the Coley group.Chapter 3 reviews the isolation, biosynthesis, biological activity, and previous syntheses of patchouli alcohol, as well as our proposal to synthesize the newly isolated natural products, 14- and 15-hydroxypatchoulol, using the developed Pd-catalyzed cascade reaction.Chapter 4 describes the total syntheses of 14- and 15-hydroxypatchoulol. It features two completed syntheses of 14-hydroxypatchoulol, each employing different gem-dichloroalkene electrophiles, completed in 9 and 12 overall steps, respectively. In addition to the completed syntheses, we report structural misassignment of 15-hydroxypatchoulol from the original isolation report.Chapter 5 reviews the isolation, biosynthesis, biological activity, and previous syntheses of plumisclerin A, along with other structurally related xenicane marine terpenoid natural products.Chapter 6 describes our synthetic studies toward plumisclerin A, in which we employ a titanocene-mediated reductive cyclization approach to construct the bicyclo[3.1.1]heptane moiety and introduce the side chain of the dihydropyran moiety through a Claisen rearrangement.