15064 research outputs found
Sort by
Self-Play Methods in Reinforcement Learning for Language Models
No full textIn this thesis we develop a series of practical algorithms for language model to self-train, by actively and strategically creating and controlling learning experiences themselves, enabling better model generalizability and training efficiency compared to traditional approaches
Dimerization and dynamics of human angiotensin-I converting enzyme revealed by cryo-EM and MD simulations
No full textAngiotensin-I converting enzyme (ACE) regulates the levels of disparate bioactive peptides, notably converting angiotensin-I to angiotensin-II and degrading amyloid beta. ACE is a heavily glycosylated dimer, containing four analogous catalytic sites, and exists in membrane-bound and soluble (sACE) forms. ACE inhibition is a frontline, FDA-approved, therapy for cardiovascular diseases yet is associated with significant side effects, including higher rates of lung cancer. To date, structural studies have been confined to individual domains or partially denatured cryo-EM structures. Here, we report the cryo-EM structure of the glycosylated full human sACE dimer. We resolved four structural states at 2.99 – 3.65 Å resolution which are primarily differentiated by varying degrees of solvent accessibility to the active sites and reveal the full dimerization interface. We also employed all-atom molecular dynamics (MD) simulations and heterogeneity analysis in cryoSPARC, cryoDRGN, and RECOVAR to elucidate the conformational dynamics of sACE and identify key regions mediating conformational change. We identify differences in the mechanisms governing the conformational dynamics of individual domains that have implications for the design of domain-specific sACE modulators.</p
First observations of solar halo gamma rays over a full solar cycle
No full textWe analyze 15 years of Fermi-LAT data and produce a detailed model of the Sun’s inverse-Compton scattering emission (solar halo), which is powered by interactions between ambient cosmic-ray electrons and positrons with sunlight. By developing a novel analysis method to analyze moving sources, we robustly detect the solar halo at energies between 31.6 MeV and 100 GeV, and angular extensions up to 45° from the Sun, providing new insight into spatial regions where there are no direct measurements of the Galactic cosmic-ray flux. The large statistical significance of our signal allows us to subdivide the data and provide the first -ray probes into the time variation and azimuthal asymmetry of the solar modulation potential, finding time-dependent changes in solar modulation both parallel and perpendicular to the ecliptic plane. Our results are consistent with (but with independent uncertainties from) local cosmic-ray measurements, unlocking new probes into astrophysical processes near the solar surface
Generalized exceptional points in nonlinear and stochastic dynamics
No full textWe study a class of bifurcations generically occurring in dynamical systems with nonmutual couplings ranging from models of coupled neurons to predator-prey systems and nonlinear oscillators. In these bifurcations, extended attractors such as limit cycles, limit tori, and strange attractors merge and split in a similar way as fixed points in a pitchfork bifurcation. We show that this merging and splitting coincide with the coalescence of covariant Lyapunov vectors with vanishing Lyapunov exponents, a feature that generalizes the exceptional points that can exist in families of non-Hermitian matrices or operators. We distinguish two classes of bifurcations associated with generalized exceptional points, corresponding respectively to continuous and discontinuous behaviors of the covariant Lyapunov vectors at the transition depending on the presence of a ℤ2 symmetry. We outline some physical consequences of this class of theories exhibiting generalized exceptional points, including nonreciprocal responses, the destruction of isochrons, and anomalous noise effects. In particular, we show that the effective diffusion coefficient on the attractor can stay finite or even diverge when the noise strength vanishes. We illustrate our results with concrete examples from neuroscience, ecology, and physics
Tensor Networks for Noninvertible Symmetries in 3+1D and Beyond
No full textTensor networks provide a natural language for noninvertible symmetries in general Hamiltonian lattice models. We use ZX-diagrams, which are tensor network presentations of quantum circuits, to define a noninvertible operator implementing the Wegner duality in 3+1D lattice ℤ2 gauge theory. The noninvertible algebra, which mixes with lattice translations, can be efficiently computed using ZX-calculus. We further deform the ℤ2 gauge theory while preserving the duality and find a model with nine exactly degenerate ground states on a torus, consistent with the Lieb-Schultz-Mattis-type constraint imposed by the symmetry. Finally, we provide a ZX-diagram presentation of the noninvertible duality operators (including noninvertible parity and reflection symmetries) of generalized Ising models based on graphs, encompassing the 1+1D Ising model, the three-spin Ising model, the Ashkin-Teller model, and the 2+1D plaquette Ising model. The mixing (or lack thereof) with spatial symmetries is understood from a unifying perspective based on graph theory
Testing the Froggatt-Nielsen mechanism with lepton flavor and number violating processes
No full textThe Froggatt-Nielsen (FN) mechanism offers an elegant explanation for the observed masses and mixings of Standard Model fermions. In this work, we systematically study FN models in the lepton sector, identifying a broad range of charge assignments (“textures”) that naturally yield viable masses and mixings for various neutrino mass generation mechanisms. Using these textures, we consider higher-dimensional operators consistent with a FN origin and find that natural realizations predict distinct patterns in lepton flavor- and number-violating observables. For Dirac and Majorana neutrinos, FN-related correlations can lead to detectable rates of charged lepton flavor violation at next-generation low-energy experiments. Majorana and type-I seesaw models predict measurable rates of neutrinoless double beta decay. Determination of inverted neutrino mass ordering would exclude the Dirac neutrino FN scenario. Only a small minority of purely leptonic FN models predict detectable flavor violation at future muon colliders, though it is possible that a combined analysis with the quark sector will reveal motivated signals. These findings highlight the power of the FN mechanism to link neutrino mass generation to testable leptonic observables, offering new pathways for the experimental exploration of lepton number and underscoring the importance of next-generation low-energy probes
CONTINUITY AND CHANGE IN SERBIAN CIVILIZATION IN THE WAKE OF THE OTTOMAN CONQUEST
No full textNo abstrac
Proof Engineering for Quantum Reasoning via E-Graphs
No full textThis dissertation is an exploration of quantum formal verification through the development of VyZX, a new formalization of the ZX-calculus. VyZX uses the categorical structure of the ZX-calculus to reason about ZX diagrams, providing a framework to establish completeness, encode common rules, and investigate applications like peephole optimization. However, working within a symmetric monoidal category introduces challenges, as the same ZX-diagram can have many equivalent representations in VyZX. This results in significant manual proof effort, as there is no single "best"; normal form, which is both tedious and difficult to fully automate. To address this, use E-Graphs, a data structure for managing equivalence classes, to streamline reasoning about equivalence. I build a system that can automatically use VyZX lemmas to prove the equivalence of two semantically equivalent diagrams if they are equal up to structure. The system performs well, solving complex associativity problems in less than a second. In benchmarks, the system is able to handle much larger problems than practical. In addition, I explore the solving of more complex equalities using custom rewrite databases. The system is able to replace and outperform Rocq’s autorewrite tactic for a common problem class in VyZX, showing its general applicability
Lenguas (latino)americanas. Entre literatura y diplomacia: vindicaciones por un universal plural. João Guimarães Rosa y Octavio Paz en la alargada Segunda Guerra Mundial (1936-1948).
No full textThis dissertation examines the intellectual and political intersections between literature and diplomacy during a decisive period of the 20th century (ca. 1935-1950). It takes as pivot of analysis the works and trajectories of two important Latin American diplomats/writers: Brazilian João Guimarães Rosa and Mexican Octavio Paz. By exploring the intricate consubstantiality of diplomacy and literature in these two authors and their respective careers and productions, the dissertation advances an interpretation of their two similar, yet different, critiques of hegemonic universalism – which they constructed as a way to articulate simultaneously (Latin)America in the 20th-century world and the world in Latin America. Through archival research and close-readings of both authors’ published and unpublished correspondences and works, the dissertation recovers the nuances of these particular universals, relocating the diplomatic and cultural role played by Mexico and Brazil within the world's complex reconfigurations between 1935 and 1950. In five chapters, the dissertation deals with: the poetic and political beginnings of Paz during the Spanish Civil War in 1937; Rosa’s important diplomatic work in Hamburg (1938-1942) at the peak of the nazi power and horror – tracing the “Germanic extension” of Rosa’s sertão or the “ sertaneja dimension” of Germany, as well as the resignification of Brazilian “cordialidade”; Rosa’s years in Hamburg as a deep questioning of metaphysics and barbarism, considering them in situ and through his obsession with language itself; Rosa’s stays in Bogotá in 1942-944 and 1948 and his views of Hispanic America and of Brazil’s continental and international ambitions in the post-war era; and, finally, Paz’s auto-reflections on Mexican condition during his diplomatic mission in San Francisco during the founding of the United Nations (1943-1945).</p
In Sync: Exploring Static and Dynamic Correlates of Parent-Child Synchrony During Puzzle Play
No full textParent–child interactions play a critical role in children’s development, yet capturing their dynamic qualities remains a methodological challenge. This dissertation introduces and evaluates the Dynamic Synchrony Scale (DSS), a novel tool for assessing moment-to-moment dyadic synchrony, and applies it to puzzle-based parent–child interactions. In Chapter 2, DSS ratings showed significant inter- and intra-rater agreement and were positively associated with ratings using an established synchrony scale, the Positive Synchrony scale (Criss et al., 2003). The DSS also predicted parents’ post-task ratings of children’s performance, an association not observed for the Positive Synchrony scale, suggesting that the DSS may align more closely with parents’ perceptions of child task performance. In Chapter 3, DSS ratings were lower during more complex puzzles, increase with children’s independent puzzle-solving skill, and vary by task order, suggesting that both task demands and dyad characteristics may relate to observed synchrony patterns. Finally, Chapter 4 used segmented DSS ratings to examine whether different forms of parent verbal input were associated with distinct synchrony patterns. Synchrony tended to rise in the moments preceding praise and verbal scaffolding, although the initial rise in synchrony prior to praise was followed by a decline, suggesting transient effects. Process-focused praise and pictorially-focused scaffolding showed the clearest pre-utterance increases, while spatially-focused scaffolding provided some evidence of a post-utterance rise, indicating the possibility of more sustained coordination following this type of scaffolding. Together, this work supports the notion of synchrony as an emergent and informative feature of dyadic interaction. By establishing the DSS as a reliable, scalable rating approach and highlighting situational elements that may foster harmonious interaction, this dissertation provides a methodological tool and contributes to a theoretical foundation for future research on when and why synchrony may reflect a positive mutual state