71446 research outputs found
Sort by
Out-of-Distribution Generalization in Biological and Artificial Intelligence.
No full textThis past decade has seen unprecedented success in Artificial Intelligence (AI), pushing the frontiers in ways most experts could have never predicted. However, most of this success has come in the form of performing well inside the data distribution the models have been trained with. Out-of-distribution (OOD) generalization still remains the Achilles’ heel of modern AI. In contrast, biological systems exhibit a remarkable ability to adapt to novel situations. This thesis addresses this critical generalization gap, by studying biological and artificial intelligence in tandem. The work presented includes new mathematical frameworks designed to better formalize generalization, behavioral benchmarks to identify the limits of both human and AI generalization capabilities, experiments to identify the underlying mechanisms driving generalization in both brains and neural networks, and engineering solutions to incorporate these findings to improve AI. To this end, this thesis presents scientific contributions made to the fields of Machine Learning, Computer Vision, Computer Graphics, Computational Neuroscience, and Psychophysics. Throughout the thesis, the goal of this work has been to advance our understanding and improve OOD generalization by working at the intersection of biological and artificial intelligence.Engineering and Applied Sciences - Computer Scienc
Addressing AI-Induced Labor Market Risks Through Enhanced Individual Capital Income
No full textWhile rapid advances in generative AI (GenAI) present significant opportunities for productivity and growth, they also risk displacing workers and deepening income inequality, particularly by increasing the relative returns to capital compared to labor. In this paper, we review current policy proposals to address these risks, noting their potential benefits as well as fiscal and implementation challenges. We complement this review with simulations estimating the fiscal impact of GenAI across OECD countries, under various unemployment and growth scenarios. We then propose a complementary policy approach: enhancing individual capital income through increased household savings and targeted financial tools. We argue that this approach can help workers manage economic transitions, reduce fiscal pressures, and promote a more equitable distribution of AI-driven capital gains. Using comparative data from OECD countries, we identify nations likely to benefit most from such policies—especially those with low household savings, limited social protection, and aging populations. We also highlight key target populations, including workers with limited access to welfare, such as the self-employed in certain countries.Version of Recor
Rewriting the Dinosaur Tail: Evidence for Dvl2 Deletion and Changes to Tailbud Development in Birds
No full textThe evolutionary reduction of the tail is one of the most striking anatomical transformations in the lineage leading from non-avian dinosaurs to modern birds. This dissertation investigates the developmental and genomic mechanisms underlying premature tail termination in birds using comparative embryology, genomics and transcriptomics in the context of the fossil record. I identify a bird-specific deletion of Disheveled-2 (Dvl2), a key conductor of Wnt signaling, that results in axial truncations when it is deleted across vertebrates. This is combined with temporally dynamic shifts in the organization and elongation dynamics of tailbud tissues in chicken relative to alligator, the closest extant relative of birds that possesses a tail. Tailbud transcriptomic analyses conserved, species-specific temporal expression dynamics of both Hox genes and Wnt ligands in both alligator and chicken. Alligator tail initiation exhibited steep bursts of expression, while chickens showed a more graded process, and alligator tailbuds showed an overall coupling of Hox and Wnt dynamics that was absent in chicken. Overall, the findings suggest a more robust onset and maintenance of elongation in alligator tail tissues, while a loss of Dvl2 and disruption of Wnt signaling in birds may have resulted in a gradual tail developmental progression that cannot sufficiently maintain progenitors of the axis.Biological and Biomedical Science
Dirty War/Dirty Loans: Bank Lending and Human Rights Violations during the 1976 -1983 Argentine Military Dictatorship
No full textThis thesis explores the relationship between international bank lending and
human rights violations during the 1976-1983 Argentine military dictatorship, commonly
known as the Dirty War. It describes the roots of the dictatorship and the regime’s
successful efforts to establish relationships with U.S., European and Asian commercial
banks, as well as the World Bank and International Monetary Fund. The regime fought a
self-described war against domestic insurgents and engaged in widespread repression,
including institutionalized torture and murder. The thesis demonstrates that lenders were
aware of the repression while they were providing financing and that they focused on
other factors to justify loan decisions, including avowed political neutrality. The thesis
also investigates the impact of various U.S. presidential administrations on the regime’s
behavior and its ability to borrow from the international banking community. Despite
differing U.S. approaches to human rights during the dictatorship, bank lending continued
without disruption as the dictatorship’s neoliberal economic policies facilitated
productive relationships with foreign lenders. An analysis of primary sources, scholarly
research and supporting data connects bank lending with the regime’s ability to remain in
power. There is substantial scholarship available on the Dirty War period, but limited
research on the impact of bank lending. The thesis offers insights into the complicity of
lenders on the perpetuation of human rights violations during the Dirty WarExtension Studie
Liquid-like dynamics in a solid-state lithium electrolyte
No full textEngineering and Applied SciencesProo
Building personalized machine learning models using real-time monitoring data to predict idiographic suicidal thoughts
No full textSuicide risk is highest immediately after psychiatric hospitalization, but the field lacks methods for identifying which patients are at greatest risk, and when. We built personalized models predicting suicidal thoughts after psychiatric hospital visits (N=89 patients), using ecological momentary assessment (EMA; average EMA responses per participant=311). We built several idiographic models, including baseline autoregressive and elastic net models (using single train/test split) and Gaussian Process (GP) models (using an iterative rolling-forward prediction method). Simple GP models provided the best prediction of suicidal urges (R2average=0.17), outperforming baseline autoregressive (R2average=0.10) and elastic net (R2average=0.06) models. Similarly, simple GP models provided the best prediction of suicidal intent (R2average=0.12) compared to autoregressive (R2average=0.08) and elastic net (R2average=0.04). Here we show that idiographic prediction of suicidal thoughts is possible, though accuracy currently is modest. Building GP models that iteratively update and learn symptom dynamics over time could provide important information to inform development of just-in-time adaptive interventions.Author's Origina
How Do Cement Firms’ Financial Outcomes Relate to Their Carbon Intensity?
No full textSustainable finance emerged as a new investment field to help investors and
governments allocate capital in line with global sustainability targets such as reversing
the trends in climate change, biodiversity loss, and resource depletion (Consolandi et al.,
2020). However, while global sustainable investments reached over $30trillion as of
2022, all global targets all global targets to reverse environmental decline have been
missed in the last several years (GSIA, 2023; IPCC, 2022).
This disconnect raises questions on the effectiveness of the approach, assumptions
and expectations sustainable finance relies on to allocate capital. Most sustainable finance
debt and capital allocations are done based on companies’ ESG ratings or similar
aggregate, subjective assessments of environmental, social and governance performance,
which recent research found ineffective in indicating real world impact (Kölbel et al.,
2020; Berg et al., 2022). The only current mainstream approach to allocate capital based
on sector specific tangible metrics is sustainability-linked instruments, which constitute
only a small portion of the overall sustainable investments. In addition to the challenges
around the tangible metrics, most investors expect similar or higher financial
performance on sustainability themed investments, and existing mainstream sustainable
research supports this expectation (Van Der Beck, 2021).
This thesis investigated the global cement sector and assessed the relationship
between financial and environmental metrics of cement companies. Carbon emission
intensity per ton of cementitious material was used as the key environmental metric as itis widely agreed as the sector’s most material environmental impact, it is a standard
metric across the sector and there are global targets set around it. Emissions intensity was
then compared against the key financial metrics of profitability, firm value and leverage.
Considering the available decarbonization levers for the cement industry, their expected
financial impact and the scale of deployment necessary to meet global cement
decarbonization targets, I tested the hypotheses that better environmental performance
will be positively correlated with companies’ leverage and valuation metrics, and
negatively correlated with its short-term profitability metrics. The results indicated no
significant statistical relationship between cement companies’ carbon intensity and any of
its key financial metrics (except for debt-to-equity leverage metric, which indicated a
negative correlation between better environmental performance and higher leverage).
These results potentially indicate that only those decarbonization investments that
are not negatively affecting the financials are being implemented on a large scale. This
might mean some decarbonization levers, even though they are commercially proven and
are expected to be implemented on a full scale to meet the global targets as part of
decarbonization pathways for the sector, are excluded or implemented only on a limited
scale across the sector, to ensure companies’ financials are not negatively affected. This
reinforces the investor expectation that the companies that have better environmental
performance than their peers should have the same or better financial returns. However,
the results from this analysis highlight that to achieve the scale of transition required to
meet the global targets and to design effective policy and financial support for sectoral
transition, it would be critical to understand what is happening as well as what is
expected to happen but is not happening
Understanding Implementation: Street-Level Bureaucrats' Resources for Reform
No full textAccepted Manuscrip
Sovereign Debt Restructuring with China at the Table: Forward Progress but Lost Decade Risk Remains
No full textSovereign debt restructuring deals have not been smooth sailing over the last few years. They have moved slowly, been marked by bickering between China and G7 stalwarts, and the outcomes have been inconsistent. Recent policy innovations, however, have successfully accelerated the pace at which deals are being completed —that’s the good news. The bad news is that China remains highly reluctant to grant permanent debt relief. Deals are coming faster, but debt relief may be insufficient to avoid repeat restructurings. This is deeply unfortunate in the post-Covid-19 context, with many lower income countries at or near debt distress (World Bank 2024 at 18).
The first part of this paper explains this recent history. We explain how the architecture of sovereign debt restructuring has evolved over the last four years in response to China’s objections to the traditional rules and procedures of the Paris Club and the IMF. After a short introductory section on China’s overseas lending boom, we review disputes over the procedures applied in the restructuring of the debt of Congo Brazzaville (2018–2019), Suriname (2021–2023), and Zambia (2020–2024). Then we review the policy innovations announced between 2022–2024 to resolve these disputes. We suggest that faster motion in recent debt restructurings is a direct result of this successful policy process, China’s pressure leading to positive procedural changes.
The latter part of this paper addresses a remaining challenge: China’s general unwillingness to grant permanent debt relief. This creates the possibility that the international restructuring architecture will be run to prefer debt maturity extension and avoid granting debt relief. This risks a repeat of the “lost decade” of the 1980s, when write-downs were systematically avoided, and many countries were trapped in serial restructurings until debt relief was finally granted in the Brady plans of the 1990s. We discuss this topic in two parts, first addressing China’s complex institutional setting, then explaining China’s policy, legal, and economic constraints to granting debt relief.
We conclude by recommending that China establishes a sovereign debt asset management company to centralize management of its problematic sovereign loans, an idea borrowed from the playbook China used in the 1990s to clean up its commercial banks. The idea would be for the new agency to take over all the problematic loans from China Development Bank (CDB) and China Eximbank. This would provide significant institutional efficiency, allow better high-level oversight, facilitate superior data management, and allow China to more flexibly respond to the exigencies of future international sovereign debt negotiations.Version of Recor
An electronic microemulsion phase emerging from a quantum crystal-to-liquid transition
No full textStrongly interacting electronic systems often exhibit a complicated phase diagram that results from the competition between different quantum ground states. One feature of these phase diagrams is the emergence of microemulsion phases, where regions of different phases self-organize across multiple length scales. The experimental characterization of these microemulsions can pose significant challenges, as the long-range Coulomb interaction microscopically mingles the competing states. Here, we observe the signatures of the microemulsion between an electronic Wigner crystal and an electron liquid in a MoSe2 monolayer using cryogenic reflectance and magneto-optical spectroscopy. We find that the transition into this microemulsion state is marked by anomalies in exciton reflectance, spin susceptibility, and umklapp scattering, establishing it as a distinct phase of electronic matter.Chemistry and Chemical BiologyPhysicsAccepted Manuscrip