Chronological feed of everything captured from Leo Laporte.
paper / leolaporte / 5d ago
DECO-MWE is a structured linguistic resource targeting Korean Multiword Expressions (MWEs) for Feature-Based Sentiment Analysis (FBSA), formalized as a Finite-State Transducer using Local Grammar Graph (LGG) methodology. Built on a cosmetics review corpus — a domain with unusually high MWE frequency — it categorizes expressions into four types: Standard Polarity, Domain-Dependent Polarity, Compound Named Entity, and Compound Feature MWEs. The resource achieves an F-measure of 0.806 on a test corpus, yielding both a reusable general-purpose polarity lexicon and a domain-adaptable finite-state methodology applicable to other NLP domains.
sentiment-analysismultiword-expressionskorean-nlplinguistic-resourcesfinite-state-transducerfeature-based-sentiment-analysiscomputational-linguistics
“DECO-MWE achieves an F-measure of 0.806 on a Korean MWE retrieval test corpus.”
paper / leolaporte / 5d ago
Multiword expressions (MWEs) represent a linguistically heterogeneous category that lacks robust, computationally useful classifications — a gap the author attributes largely to poor feature selection. Laporte argues that not all available features are equally reliable for assigning MWEs to classes, and that feature quality directly determines the downstream utility of any resulting classification. The paper proposes an enhanced classification framework designed with cross-linguistic coverage in mind, drawing on prior work across multiple languages to improve generalizability.
multiword-expressionscomputational-linguisticsnlptext-classificationfeature-selectionmultilinguallexical-semantics
“MWEs are a heterogeneous category with a significant unmet need for principled classifications.”
paper / leolaporte / 22d ago / failed
paper / leolaporte / Apr 26
A robust method using subgiant star ages dates the Gaia-Sausage-Enceladus (GSE) merger to ~11 Gyr ago, coinciding with the Tainá starburst at 11.2 ± 0.1 Gyr that birthed coeval in-situ globular clusters (GCs). GSE's metal-rich GCs formed at 10.9 ± 0.1 Gyr during merger interactions, with ω Centauri as the likely surviving core, its stars matching ages and metallicities while showing bar resonance effects. Kinematic transitions at [Fe/H] ~ -1.33 and proto-MW GCs with disc-like orbits up to 13.0 ± 0.5 Gyr old indicate disc formation began at z_disc_form ≳ 4, pre-merger.
milky-way-mergergaia-sausage-enceladusglobular-clustersgalactic-evolutionomega-centauristellar-agesdisk-formation
“The last significant Milky Way merger (GSE) occurred ~11 Gyr ago.”
tweet / @leolaporte / Apr 20 / failed
My Signal address is https://signal.me/leolaporte.24
paper / leolaporte / Apr 11
Using basis function expansions applied to a high-resolution N-body simulation of the LMC-SMC system in isolation, this study quantifies the mutual dark matter halo distortions of the Magellanic Clouds prior to Milky Way infall. The SMC induces a ~20 kpc dynamical friction wake and dual overdensities in the LMC halo at ~60 and ~100 kpc, while itself losing two-thirds of its initial dark matter mass to the LMC by infall. Critically, these perturbations persist across multiple SMC pericenters and produce a highly asymmetric acceleration field, meaning static or spherically symmetric halo models are insufficient for accurate orbit integration. The authors conclude that 1:10 mass-ratio encounters generate characteristic, scale-invariant halo deformations — a result with direct implications for merger rate estimates and dark matter model constraints.
dark-mattergalaxy-dynamicsn-body-simulationmagellanic-cloudsgravitational-interactionsastrophysicsbasis-function-expansions
“The SMC has lost approximately two-thirds of its initial dark matter mass to the LMC by the time of Milky Way infall.”
paper / leolaporte / Apr 11
This paper presents a patient-specific, multi-scale computational lung model that integrates CT-derived airway geometry with algorithmically generated smaller airways to simulate ventilation dynamics. Tissue mechanics are modeled via nonlinear elasticity coupled with fluid dynamic pressure within the bronchial tree, with airflow accounting for both inertia and static airway compliance. Finite element simulations are used to resolve spatio-temporal distributions of airflow and wall shear stress across the full lung architecture. The framework enables physiologically grounded investigation of ventilation heterogeneity in personalized lung models.
computational-biologylung-modelingfluid-dynamicsbiomechanicsfinite-element-analysismedical-imagingrespiratory-physiology
“Large airway geometry and lung envelope are derived directly from patient CT scans, enabling personalized modeling.”
paper / leolaporte / Apr 11
Stepped-wedge cluster randomised trials (SW-CRTs) pose analytical challenges when composite endpoints are evaluated using generalized pairwise comparisons (GPC), as most estimators fail to adequately account for clustering and temporal trends. A comprehensive simulation study across varying ICCs, cluster autocorrelation coefficients (CAC), and treatment effect sizes found that most GPC approaches inflate Type I error. Only two methods — a hierarchical mixed-effects model with sequence and cluster-level random slopes (b4) and a cluster-restricted probabilistic index model (c2) — consistently maintained nominal error rates. Between the two, c2 demonstrated superior statistical efficiency, particularly under strong clustering, low CAC, or temporal trends, while both converged in performance for large treatment effects.
clinical-trialsstepped-wedge-designcluster-randomised-trialssimulation-studygeneralized-pairwise-comparisonscomposite-endpointsbiostatistics
“Most GPC-based estimators that ignore clustering or time effects fail to control Type I error in SW-CRTs.”
paper / leolaporte / Apr 11
Using JWST NIRSpec-IFU high-resolution spectroscopy, Maiolino et al. confirm a HeII λ1640 emitter at z=10.6, located just 3 physical kpc from the well-known galaxy GN-z11. The source shows no detectable metal lines and an exceptionally high HeII equivalent width (>20 Å), with the emission spectrally resolved into two components separated by 120 km/s. The authors systematically rule out alternative ionization mechanisms and conclude that Population III stars — the universe's first, chemically pristine stellar generation — represent the most plausible explanation, marking a significant step toward the first observational confirmation of Pop III star formation.
population-iii-starsearly-universejwstspectroscopygalaxy-formationcosmologyhigh-redshift
“A HeII λ1640 emitter at z=10.6 is confirmed at 3 physical kpc from galaxy GN-z11, corroborating a prior medium-resolution detection.”
paper / leolaporte / Apr 11
The Vera C. Rubin Observatory has released Data Preview 1 (DP1), its inaugural public dataset derived from 1,792 commissioning exposures taken over 48 nights in late 2024 using LSSTComCam on Cerro Pachón, Chile. Covering ~15 deg² across seven fields in six photometric bands (ugrizy), DP1 delivers coadded 5σ point-source depths reaching g=26.18 and r=25.96 in the deepest field, with median PSF FWHM of 1.14 arcseconds. The 3.5 TB dataset catalogs ~2.3 million astrophysical objects and 93 newly discovered solar system objects, and is accessible to data rights holders via the cloud-based Rubin Science Platform ahead of full LSST operations in 2026.
astronomytelescope-observatorysky-surveyoptical-imagingastrophysicsdata-releasescientific-instrumentation
“DP1 covers approximately 15 deg² across seven non-contiguous fields, each observed in six broad photometric bands (ugrizy), based on 1,792 exposures over 48 nights in late 2024.”
tweet / @leolaporte / Feb 25
A 17-hour outage of the hosting provider Megaphone triggered a metadata or entitlement error within Apple Podcasts, incorrectly restricting free content to paid access. Service has since been restored for MacBreak Weekly listeners.
podcast-updateservice-disruptiontechnical-issuesapple-podcastsmegaphone-outageleo-laporte
“Megaphone, a podcast hosting provider, experienced a 17-hour outage.”
tweet / @leolaporte / Jan 4
Leo Laporte utilized the Keybase platform to publicly verify ownership of his X (formerly Twitter) account. This process involved linking his X profile to his Keybase identity, thereby leveraging Keybase's cryptographic proof system to establish a verifiable connection between the two online presences. This method provides a decentralized and secure way to confirm digital identities.
social-media-verificationkeybasetech-identityonline-security
“Leo Laporte's X account ownership is verified.”