FAS Theses and Dissertations

Life is full of stress-inducing activities, which have the potential to shape habitual behaviors that can increase anxiety and decrease overall well-being if not dealt with properly. While existing stress management technologies aim to mitigate these behaviors, such interventions are largely impersonal and have a high barrier to entry, offering generalized guidance and requiring users to manually trigger the intervention. This thesis presents a novel, smartwatch-based system for disrupting stress habits that automatically detects moments of physiological arousal and delivers personalized, in-the-moment reflection exercises.

A within-subjects user study examined 24 participants engaging with a prototype of this system, finding that 87.5% reported greater awareness of their stress habits after using the application. Participants also significantly preferred (p = 0.003) the automatic triggering over scheduled methods. Ultimately, this work contributes a novel integration of physiological sensing and behavior-change theory, offering insights into how wearables can support emotional regulation and habit-loop disruption within everyday routines.

Economic mechanism design shapes our economic and social systems, silently influencing outcomes for billions worldwide—from auctions powering global online advertising to matching algorithms determining hospital placements for medical residents. Yet, despite decades of significant theoretical advances, classical economic approaches have encountered analytical bottlenecks, leaving fundamental questions unresolved. Inspired by transformative breakthroughs in the application of AI and machine learning to the natural and physical sciences, this thesis pioneers and extends Differentiable Economics, a computational framework that integrates economic theory with deep learning to systematically design optimal economic mechanisms.

Differentiable Economics reformulates economic design as an end-to-end optimization problem, representing auctions, data markets, and matching mechanisms through differentiable neural architectures. This allows previously intractable economic problems to be solved systematically and with practical computational and statistical efficiency. This thesis is organized into three main parts:

In Part 1, Auctions, I propose two neural architectures: RegretNet, a characterization-free method that flexibly addresses the design of complex multi-item auctions, and RochetNet, a characterization-based architecture that explicitly enforces incentive compatibility for single buyer settings. These architectures successfully replicate known optimal solutions, validate important conjectures, and discover entirely new  mechanisms in scenarios where analytical solutions remain elusive. Additionally, I extend the RochetNet framework to a sequential setting by developing a new reinforcement learning approach that outperforms traditional RL methods and analytical baselines.

In Part 2, Data Markets, I design neural network architectures capable of learning signaling schemes for selling information. This framework can effectively handle new economic constraints arising from obedience conditions, modeling the action that an agent will take upon receiving information, and incentive compatibility. This approach not only replicates established theoretical benchmarks but also identifies and validates novel optimal signaling strategies previously unknown in economic theory.

In Part 3, Matching Markets, I introduce permutation-equivariant convolutional neural networks alongside novel differentiable surrogate metrics for stability and incentive compatibility in the design of two-sided matching markets. This combination enables the computational exploration and characterization of previously unknown trade-offs between stability and incentive compatibility.

Taken together, these contributions highlight the potential of Differentiable Economics as a flexible and powerful methodology for economic design. This thesis takes foundational steps towards democratizing economic mechanism design, providing economists and computer scientists alike with new, accessible tools to address long-standing theoretical challenges in optimal economic design.

Large language models hallucinate false content and produce sycophantic outputs, yet the dominant approaches to these failures, such as Reinforcement Learning from Human Feedback, Constitutional AI, and fine-tuning, share a common omission: they target symptoms without first asking what function an LLM serves. This thesis begins with that prior question. I argue that a central class of LLM outputs functions as assertions, propositions put forward as true, and should therefore be governed by assertoric norms. Operationalizing Timothy Williamson’s Knowledge Norm of Assertion for mechanistic systems, I define the Epistemic Governance Norm of Assertion (EGNA), which requires (i) that a system possess the capacity to partition epistemically available from unavailable content, and (ii) that this partition govern what is asserted. Drawing on recent mechanistic interpretability research, I show that modern LLMs already approximate the first condition. Current training objectives, however, structurally decouple epistemic registration from assertoric output, producing hallucination and sycophancy as predictable consequences. I implement a two-stage EGNA scaffold enforcing epistemic classification as a precondition of assertion and evaluate it across five frontier models. The scaffold reduces hallucination to zero or near-zero rates across all models and substantially suppresses sycophancy, while maintaining competitive response rates on legitimate queries. A control scaffold preserving the two-stage structure without the epistemic partition leaves hallucination largely unchecked, isolating the partition as the operative mechanism. The results also reveal that social inference enters epistemic classification itself, a failure mode that purely behavioral interventions would not have surfaced. Ultimately, getting the prior question right is not a philosophical exercise but a precondition for these systems to perform their function well.

LLMs have gained rapid popularity in recent years, with the ability to align outputs to human preferences and objectives. Eager end users have caused surging demand for LLM inference. This has motivated new research to improve efficiency, reduce costs, and better understand systems in inference serving, alongside decades of similar research in distributed training. Meanwhile, the wave of systems- and hardware-oriented reinforcement learning (RL) research has just begun. RL offers a unique paradigm that combines inference, preparation, and training phases. Although RL has been a popular arena for algorithmic inventions, there is still a need to better understand how RL policies and phase workloads affect underlying systems, beyond algorithms. This requires carefully designed experiments that vary key axes to test system effects.

To address this gap, this thesis conducts experiments along four axes: model family/policy, reward paradigm, task/dataset, and scaling. We build profiling extensions atop VeRL and collect metrics such as power, latency, energy, and other system-level metrics at the granularity of phases and subphases—a novel contribution in RL. We find that different RL policies induce distinct phase signatures, driven both by time spent in phases and by extra computational machinery. We quantify the costs of different reward-scoring paradigms—a learned reward model or an explicit reward function—in data movement and extra computation. We find that different tasks and datasets alter the response-length structure and rollout imbalance, contributing to latency differences. Finally, we outline the best use cases for different scaling configurations on H200 and A100 GPUs, which are strongly dependent on phase and stability requirements. We hope that our work may motivate new RL algorithms that address the less desirable attributes of phases, advance asynchronous RL methods that lower transition costs between phases, and inform data centers and cloud operators about the distinct workloads of RL. Overall, we hope that the profiling suite and metrics pipeline will inspire further energy-aware innovation in RL.

Research and development (R&D) competition generates innovation but can also produce socially wasteful duplication of effort. In our thesis, we model this dynamic through a stochastic all-pay auction, where firms make sunk-cost investments to increase their chances of obtaining a technological prize.

Equilibrium and welfare properties are then analyzed for deterministic all-pay auctions, proportional (Chinese) auctions, and a broader class of stochastic “share-of-score’’ mechanisms. The analysis derives equilibrium bidding behavior, welfare bounds, and approximate Price of Anarchy guarantees, and identifies conditions under which total social welfare asymptotically increases with greater competition and more firms.

Policy interventions are then studied from the perspective of a social planner seeking to reduce inefficient investment while preserving innovation incentives. Transfers, predictive information about firm valuations, and incentive-compatible reporting mechanisms are examined as coordination tools.

Finally, computational methods — including multi-agent reinforcement learning, numerical approximation, and differentiable mechanism design — are used to approximate equilibria and explore welfare-improving mechanisms in environments without closed-form solutions.

Taken together, these results provide the crucial theoretical and computational tools for understanding how institutional design shapes incentives and efficiency in R&D, providing hope for a more socially-optimal future allocation of scientific resources and investment.

Many networks contain multiple views of the data, which occur when you have one set of vertices with multiple sets of edges that connect them. These multiple sets of edges represent different sets of relationships between the vertices and can be a potent source of information about the underlying groupings of the nodes. Graph Neural Networks (GNNs) are often used to extract information from multi-view data, however, the mechanisms underlying the effectiveness of GNNs are not well understood. We analyzed the performance of two different frameworks for simple GNNs on multi-layer data generated using the multi-layer contextual stochastic block model, which allowed for fine-grained adjustment of graph properties. We found that the layer-wise probability of a community difference and the graph signal-to-noise ratio in each layer both affect model performance, and present intuition behind why this is the case.

APO is a two-stage reinforcement learning algorithm that is used to fine-tune large language models (LLM) on specific tasks. Stage 1 estimates the KL-regularized optimal value function V offline, and Stage~2 trains the policy online by regressing on optimal advantage targets. The log-mean-exp (LME) estimator of V* used in Stage 1 is downward-biased due to Jensen's inequality. This bias is significant, especially when the budget is small and the pass rate is low.

This thesis develops and evaluates two families of alternative estimators. Firstly, inspired by the replica trick from statistical physics, we develop two types of replica-based estimators. We prove that the single-replica plug-in estimator does not reduce bias under a fixed budget, and develop the multi-n slope estimator, which exploits the structure of the replica trick more directly. Secondly, for the binary reward, we show that LME is equivalent to plugging the maximum-likelihood pass-rate estimate into a known closed form, which causes endpoint snapping when the pass rate is low. Inspired by conjugate priors, we develop two different Beta-smooth estimators to solve this problem.

We evaluate these estimators in three experiments: (1) a scalar experiment where pass rate is fixed, (2) a contextual bandit experiment with 500 prompts of different pass rates, and (3) a full two-stage simulation of the APO pipeline. We discovered that replica-based estimators have a lower RMSE compared to LME when the pass rate is very low. However, they have a much larger advantage distortion that limits their practical value in Stage 2. The Beta-smooth estimators perform better compared to LME in all three experiments. In particular, the Jeffreys Beta-smooth estimator (p = (k+0.5)/(N+1)) achieves the best RMSE, reduces advantage distortion compared to LME, and carries over the improvement in Stage 1 into 25% lower parameter variance in the learned policy in Stage 2. We recommend it as a drop-in replacement for LME in APO's Stage 1.

Humanity is at a precipice: democracy is declining from a historical peak as backsliding occurs around the world, while artificial intelligence is emerging as a revolutionary technology that could permanently upend the socio-economic fabric underpinning modern society. This thesis presents four pieces of original research as part of a collective attempt to build upon our understanding of democratic processes, especially as democracies around the world adapt to the age of AI. In the following chapters, we (1) measure persistent non-response bias in a well-calibrated nationwide poll for the US Presidential Election, finding that the five largest states saw a >99% reduction in effective sample size, (2) extend a theoretical measure of polarization to participatory budgeting, finding that a newly proposed measure degenerates to a well-known measure ~99.4% of the time when applied to real elections, (3) adapt proportional veto core to the setting with infinite candidates (heralded by generative AI) as a measure of bridging, finding a strong pattern that some voting rules consistently perform better in bridging, and (4) demonstrate that innocuous-looking datasets that are easy to produce can poison AI models to have partisan leanings even after heavy semantic filtering, and develop attribution techniques that reduce subliminal effects by 100% when used for filtering.

High-precision robotic tasks, such as peg insertion, pose a challenge for learned world models. The critical states that determine success occur within an extremely narrow portion of each task trajectory, causing uniform sampling to underrepresent them. This thesis investigates whether targeted data selection improves the generation quality and downstream success of a 4D latent world model on the ManiSkill3 PegInsertionSide-v1 benchmark. We propose \textit{Pre-insertion Keyframe Sampling}, an algorithm that identifies and injects a single geometrically aligned pre-insertion frame into training data, and introduce a \textit{Peg-Hole Alignment Metric} to evaluate 3D generation quality beyond standard image metrics. Our results demonstrate that while the intervention yields statistically significant improvements in translational alignment, these gains do not translate to higher task success rates. A ground-truth subgoal ablation reveals a 56-percentage-point performance gap, identifying the structural fidelity of 3D generations as the primary bottleneck. This suggests that while data coverage improves spatial precision, high-precision control requires increasing the geometric resolution and structural integrity of the world model's output.

This thesis analyzes the software architecture, development decisions, and testing procedures of the Harvard Undergraduate CubeSat (HUCSat), a six-year student-led nanosatellite project dedicated to evaluating the economic and practical viability of nitinol shape memory alloys to rotate solar panels. Section 1 provides a contextual background concerning the HUCSat’s overall hardware design and layout of circuit boards. Section 2 dives specifically into the Flight Controller Board (FCB) chip, defending how our team progressed through rapid prototypes to our ultimate settlement on the RP2350 microprocessor. Section 3 presents all the major software files that the FCB employs to schedule and operate all necessary tasks. This includes a main loop file, data processing files, finite-state machine (FSM) files, drivers, managerial classes, and configuration files. Some of this work is novel, developed for this thesis from scratch or re-purposed from open-source development code. Across sections 1-3, I detail the design considerations that had to be considered, justifying the transition from one variation to the next. Finally, Sections 4-5 underscore the testing and experiment procedure to ensure robustness and clarity of the mission, respectively.

The primary contributions of this thesis are threefold. First, I present the implementation of new low-level software components, including hardware drivers and device managers abstractions tailored to the RP2350 PROVES Kit open-source flight controller. Second, I defend the development of testing methodologies (covering over 1,100 lines of newly-written test code), operational procedures, and experimental designs used to validate system behavior. Overall, such details will underscore the satellite’s resilience to concerns including radiation-induced faults, Low-Earth Orbit (LOE) temperature fluctuations, and memory constraints. Third, for the FCB, I introduce a novel dynamic, fault-tolerant, and non-blocking task-scheduling framework to conduct all HUCSat operations. Specifically, this framework integrates a system-wide finite state machine and data processing class with asynchronous task execution and command-data handling extensions to enable adaptive-decision making based on individual component health and telecommand data. In addition, its usage of a radiation-tolerant watchdog timer and non-volatile reboot and counter registers that have access to backup power systems ensure dynamic fault-tolerance. In sum, these contributions aim to provide a concrete reference to future generations constructing CubeSat flight software. To the best of knowledge, this is Harvard College’s first ever nanosatellite launch supported by NASA’s CubeSat Launch Initiative (CSLI) initiative.