Rediscovery benchmark

Integrating temporal regularity metrics from synchronized LSTM-extracted emotion profiles of facial and vocal signals into a multi-modal DNN fusion architecture will significantly improve PHQ-8 depression severity prediction compared to static multi-modal feature fusion.

IV: Integration of synchronized LSTM-AE temporal emotion profile features into the multi-modal DNN fusion pipeline · DV: Mean Absolute Error (MAE) of PHQ-8 depression severity prediction

Mechanism: Depression severity disrupts the natural micro-fluctuations and temporal dynamics of affective expression across facial and vocal channels. By capturing these sequential disruptions via LSTM-AE bottleneck features, severity-dependent temporal signatures are preserved that static frame-based extraction discards. Fusing these dynamic temporal features allows the DNN to learn non-linear mappings between affective temporal degradation and clinical severity, thereby improving prediction accuracy over static multi-modal baselines.

grounded on: R138876, R138879, R138934, R138931, R139009

Methylated β-cyclodextrins (DM-β-CD and HP-β-CD) enhance the transmucosal apparent permeability of both hydrophilic peptides and small-molecule drugs by a comparable 2.5–4.0 fold increase, independent of the drug’s lipophilicity, due to reversible apical membrane fluidization rather than solubility enhancement alone.

IV: Cyclodextrin complexation (2% w/v DM-β-CD or HP-β-CD vs. equivalent free-drug control) · DV: Apparent permeability coefficient (Papp) across respiratory and ocular mucosal barriers

Mechanism: The hydrophobic methyl/HP substituents on the cyclodextrin ring insert into the apical lipid bilayer, disrupting acyl chain packing and transiently increasing membrane fluidity. This lowers the activation energy for both transcellular diffusion of lipophilic drugs and solvent-filled pore passage of hydrophilic peptides, decoupling permeation enhancement from aqueous solubility gains while preserving histological integrity.

grounded on: R155599, R155456, R151616, R155499, R151525, R151621

The maximum magnitude of resistance evolution during experimental adaptation is inversely constrained by the biophysical complexity of the drug target, such that single-target antibiotics yield ≥100-fold MIC increases while multi-target membrane disruptors cap at ≤5-fold.

IV: Drug target complexity (single enzymatic target vs. multi-component membrane structure) · DV: Maximum achievable MIC fold-increase

Mechanism: Resistance to single enzymatic targets (e.g., DNA gyrase) requires only high-affinity point mutations that drastically reduce drug binding without compromising cellular function, permitting exponential MIC jumps. In contrast, resistance to membrane-targeting agents requires coordinated, global alterations to membrane charge, fluidity, and permeability to maintain homeostasis, which is biophysically constrained and metabolically costly, imposing a low ceiling on resistance magnitude.

grounded on: R1351005, R1351027

Transition metal dichalcogenide (TMDC)-based field-effect transistor (FET) gas sensors achieve a limit of detection at least two orders of magnitude lower than TMDC-based chemiresistors for ammonia detection at room temperature, due to the electrostatic gating mechanism that modulates the 2D channel conductivity in response to surface adsorbates, amplifying the transduction signal compared to simple two-terminal resistive changes.

IV: Sensor transduction architecture (Field-effect transistor vs. Chemiresistor) applied to TMDC sensing layers · DV: Limit of detection (LOD) for ammonia (NH3)

Mechanism: The FET architecture applies a perpendicular electric field that depletes or accumulates carriers in the atomically thin TMDC channel. When NH3 molecules adsorb and donate electrons, they shift the threshold voltage, multiplicatively amplifying the channel current change relative to the number of adsorbed molecules. In contrast, chemiresistors measure only the direct parallel resistance change of the surface film, which is attenuated by bulk scattering, grain boundary effects, and contact resistance, yielding a weaker signal per adsorbate event.

grounded on: R139328, R139336

Substituting metallic Pt nanoparticles with Ni-CH3CH2NH2 molecular complexes on C3N4/KNbO3 heterojunctions will increase CO2 photoreduction yield by 3.5–5.8 fold while suppressing H2 evolution, due to Pt's preferential proton reduction kinetics outcompeting CO2 activation on niobate surfaces.

IV: Co-catalyst identity on C3N4/KNbO3 (Pt nanoparticles vs. Ni-CH3CH2NH2 molecular complex) · DV: CO2 photoreduction rate (μmol h-1 g-1) and CH4/CH3OH product selectivity ratio

Mechanism: Pt nanoparticles possess low hydrogen evolution overpotential and strong H-adsorption energy, kinetically diverting photogenerated electrons from the KNbO3 conduction band to proton reduction rather than CO2 activation, as evidenced by the near-absent CH4 yield on Pt-g-C3N4/KNbO3. Ni-CH3CH2NH2 complexes provide localized Lewis acid sites that stabilize CO2•− radical intermediates while sterically hindering H-atom coupling, redirecting charge carriers toward multi-electron CO2 reduction pathways that the niobate lattice intrinsically supports. The organic co-catalyst maintains structural integrity on niobate surfaces under irradiation, as demonstrated by its high activity in H2 evolution systems, but its electronic coupling with the niobate CB is better matched to CO2 reduction thermodynamics than to proton reduction.

grounded on: R46233, R46231, R46221, R46156

In B,N-codoped TiO2 nanoparticles, boron incorporation creates shallow electron trap states that reduce the trapped electron relaxation time from the intrinsic 500 ps baseline to under 200 ps, which proportionally increases the apparent quantum efficiency of visible-light-driven hydrogen evolution from water splitting by 1.5- to 2.0-fold relative to nitrogen-doped TiO2.

IV: Boron dopant concentration in the N-doped TiO2 lattice · DV: Apparent quantum yield of visible-light-driven H2 evolution

Mechanism: R45116 establishes that trapped electrons in TiO2 nanoparticles undergo a 500 ps relaxation process. R46074 demonstrates that B,N-codoping activates visible-light-driven H2 evolution. Boron substitution is hypothesized to introduce shallow trap states that accelerate electron release from deep traps (shortening the 500 ps relaxation to <200 ps), thereby reducing bulk electron-hole recombination and increasing the flux of conduction-band electrons available for proton reduction at catalytic sites.

grounded on: R45116, R46074, R46123

Integrating exception-handling structures into goal models reduces decision latency for requirements engineers resolving ambiguous regulatory compliance requirements under uncertainty.

IV: Goal model representation type (standard vs. exception-handling-augmented) · DV: Decision latency (time to resolve ambiguous regulatory compliance points)

Mechanism: Exception-handling structures in goal models explicitly map potential failure states and regulatory ambiguities, reducing the cognitive search space and providing structured resolution paths. This directly supports early decision-making under uncertainty by pre-defining exception branches, thereby accelerating the requirements engineer's ability to classify and resolve ambiguous regulatory claims without iterative backtracking.

grounded on: R211145, R211137, R211198

Integrating semantic-aware local-global attention mechanisms into mobile-friendly vision transformers significantly improves episodic linear probe performance on long-tailed visual recognition benchmarks compared to standard global self-attention architectures.

IV: Attention module design (semantic-aware local-global fusion vs. standard global self-attention) · DV: Tail-class recall rate during episodic linear probing

Mechanism: Semantic-aware local-global attention captures fine-grained local textures and contextual global semantics simultaneously, producing richer feature embeddings that mitigate head-tail distribution skew. Standard global self-attention tends to accumulate dominant head-class gradients, suppressing tail-class discriminability. When paired with episodic linear probing, which adaptively selects representative samples per class, the enriched local-global features allow the probe to better isolate and classify underrepresented tail instances that global-only attention obscures.

grounded on: R1856122, R1855993, R1856094

Integrating Context Autoencoder pre-training into Active Token Mixer backbones for semantic segmentation will yield superior mean Intersection over Union on ADE20k compared to standard Masked Autoencoder pre-training, by aligning self-supervised contextual learning with the architecture's token-mixing inductive bias.

IV: Pre-training objective (Context Autoencoder vs. Masked Autoencoder) · DV: Mean Intersection over Union (mIoU) on ADE20k

Mechanism: The Context Autoencoder's objective explicitly models long-range contextual dependencies, which synergizes with ActiveMLP's active token mixing strategy to produce richer feature representations that transfer more effectively to dense prediction tasks.

grounded on: R1802991, R1803091

In Diptera and Lepidoptera, Neotropical lineages exhibit a significantly higher detection rate of cryptic diversity using distance-based MOTU delimitation methods (Barcoding gap, NJ clustering) compared to model-based coalescent methods (GMYC, BINs), whereas Palearctic lineages show the inverse pattern, favoring model-based methods for species discovery.

IV: Biogeographical region (Neotropical vs. Palearctic) · DV: Relative efficacy of MOTU delimitation methods (distance-based vs. model-based) in exceeding current taxonomy species counts

Mechanism: Neotropical lineages likely harbor deeper, older cryptic divergences shaped by long-term climatic stability and Pleistocene refugia, which are better captured by distance-based thresholds that reflect cumulative genetic divergence. In contrast, Palearctic lineages may have undergone more recent, rapid post-glacial radiations or extensive introgression, generating gene tree discordance that is more accurately resolved by coalescent-based model-based methods (GMYC, BINs) which account for ancestral polymorphism and incomplete lineage sorting.

grounded on: R146639, R145434, R145304, R137111, R145495, R145482

Coupling Ocean Biogeo Chemistry with Sea Ice dynamics in CMIP5 Earth System Models reduces the projected summer Arctic sea ice extent decline by 18% (95% CI: 13-23%) relative to physical-only ocean configurations under RCP8.5 forcing, due to biogeochemically-mediated changes in upper-ocean stratification and solar radiation penetration.

IV: Inclusion of an active Ocean Biogeo Chemistry module in coupled ocean-sea ice configurations · DV: Rate of summer Arctic sea ice extent decline

Mechanism: Ocean biogeochemical processes (e.g., dissolved organic matter absorption, biological carbon pump) alter upper-ocean density stratification and increase solar radiation attenuation in the mixed layer. This reduces summer heat uptake by the ocean surface, lowering upward turbulent heat flux to the atmosphere and thereby slowing sea ice melt rates.

grounded on: R23287, R23471, R23408, R23326, R23368, R23300

Coupling Ocean Biogeo Chemistry modules with Sea Ice dynamics in Earth System Models amplifies the radiative forcing response to Aerosol perturbations by 20-30% compared to physically coupled models lacking biogeochemical cycles, mediated by sea ice-albedo and carbon cycle feedbacks.

IV: Aerosol optical depth forcing · DV: Net top-of-atmosphere radiative imbalance and surface primary productivity

Mechanism: Increased aerosol optical depth scatters incoming solar radiation, cooling the lower atmosphere and ocean surface. This surface cooling expands sea ice extent, which increases surface albedo and thermally isolates the ocean, reducing heat and gas exchange. Reduced light penetration and altered nutrient stratification under expanded sea ice suppress phytoplankton growth (Ocean Biogeo Chemistry). Diminished biological activity decreases dimethyl sulfide (DMS) emissions and CO2 outgassing, potentially modifying cloud condensation nuclei concentrations and atmospheric chemistry, which further amplifies the initial aerosol radiative effect, creating a positive feedback loop that magnifies the climate response.

grounded on: R23368, R23300, R23326

Co-doping ALD-grown Al2O3 films with La3+ at 2–5 at.% will increase the Er3+ electroluminescence external quantum efficiency by 25–40% relative to Yb/Er-only co-doped controls, because La3+ incorporation passivates non-radiative defect sites that disrupt the Yb3+→Er3+ resonant energy transfer pathway.

IV: La3+ dopant concentration (at.%) introduced during ALD precursor cycling in Al2O3:Yb,Er films · DV: Er3+ electroluminescence external quantum efficiency (EQE) at 1.54 μm

Mechanism: La3+ substitutes into octahedral Al3+ sites in the Al2O3 matrix, locally compensating charge imbalance and reducing the density of sub-bandgap defect states. This extends the exciton diffusion length, thereby increasing the probability that Yb3+ excitation energy reaches Er3+ ions before non-radiative recombination, directly boosting the electroluminescent output.

grounded on: R1469756, R1469867, R1469781, R1469850

Precursor ligand class systematically dictates the saturation growth-per-cycle (GPC) ceiling in rare-earth oxide ALD, with formamidinate ligands enabling >2.5× higher GPC than cyclopentadienyl (Cp) ligands, and β-diketonate (thd) ligands yielding the lowest GPC, due to ligand-specific surface saturation coverage.

IV: Precursor ligand class (formamidinate vs. cyclopentadienyl vs. β-diketonate/thd) · DV: Growth per cycle (GPC) in angstroms

Mechanism: The steric footprint and surface saturation density of the precursor ligand directly govern the mass deposited per ALD cycle. Formamidinate ligands (R1470333) are planar and less sterically demanding than bulky cyclopentadienyl rings (R1470296, R1470149) or chelating β-diketonate groups (R1470264, R1470236), allowing a higher density of metal centers to adsorb per unit surface area. This geometric difference translates to a predictable, ligand-class-dependent scaling of GPC across diverse rare-earth targets, independent of the specific metal center.

grounded on: R1470333, R1470296, R1470264, R1470239, R1470149

In water-oxidant ALD of transition metal oxides, increasing substrate temperature within the 150–250 °C window induces a compensatory trade-off where a decline in growth-per-cycle (GPC) is offset by a disproportionate increase in atomic packing density, yielding a net positive correlation between temperature and areal mass deposited per cycle.

IV: Substrate deposition temperature (150 °C to 250 °C) · DV: Net areal mass deposited per cycle (product of GPC and volumetric density)

Mechanism: Elevated temperatures within the ALD window reduce surface hydroxyl coverage and accelerate organic ligand desorption. This decreases the number of available nucleation sites per precursor pulse (lowering GPC), but simultaneously drives surface reconstruction, eliminates sub-oxide/porous phases, and increases intrinsic film compactness. The densification effect dominates the GPC reduction, yielding higher areal mass per cycle despite fewer atoms being incorporated per pulse.

grounded on: R676130, R676169, R676153, R676159

Deep learning configurations pairing encoder-centric segmentation architectures (U-Net family) with single-institution private CT datasets will exhibit a performance profile of sensitivity >95% and specificity <95%, whereas configurations pairing pre-trained classification backbones (ResNet-50/COVNet) with multi-center heterogeneous datasets will exhibit sensitivity 80–90% and specificity >90%, reflecting a systematic architecture-data source interaction effect on the sensitivity-specificity trade-off.

IV: Architecture-data source configuration (encoder-centric segmentation on private/single-site data vs. pre-trained classification backbone on multi-center/heterogeneous data) · DV: Model sensitivity and specificity percentages on held-out test sets

Mechanism: Encoder-centric architectures (U-Net family) optimize local receptive fields for lesion boundary delineation, maximizing true positive rates on homogeneous training distributions but lacking global semantic regularization, which limits specificity generalization. Pre-trained classification backbones (ResNet-50) transfer robust global feature representations that regularize decision boundaries across heterogeneous scanner protocols and patient demographics, trading marginal sensitivity for stabilized specificity.

grounded on: R700920, R700923, R675126, R700955, R700959

In deep learning-based event camera depth estimation, increasing the temporal bin count (B) used to rasterize asynchronous event streams improves accuracy only up to a saturation threshold, beyond which performance plateaus due to the combined effect of fixed-window timestamp discarding and noise accumulation in high-B event surfaces.

IV: Temporal quantization resolution (number of bins, B) used to convert asynchronous event camera streams into dense tensors for deep neural network processing · DV: Depth estimation accuracy, quantified as absolute relative error

Mechanism: Event cameras output asynchronous spikes that require conversion to dense representations for CNN processing via temporal quantization into B bins (R642467, R642470). Architectures that retain only recent timestamps and discard prior ones (R642476) impose a fixed temporal integration window. As B increases, finer intra-window motion parallax is resolved, improving depth estimation for stereo/monocular tasks (R1411261, R1411264, R1411295, R1411274, R1411281). However, beyond a critical B, additional bins encode redundant or noisy events within the fixed window rather than extending the effective temporal baseline, causing accuracy gains to plateau.

grounded on: R642467, R642470, R642476, R1411261, R1411264, R1411295, R1411274, R1411281

Coarsening the temporal resolution of onshore wind speed time series systematically reduces the estimated wind power density derived from fitted Weibull distributions, with the magnitude of underestimation inversely scaling with total time series length.

IV: Temporal resolution of wind speed measurements (e.g., 10-minute vs 1-hour averaging intervals) · DV: Estimated wind power density (W/m²) and Weibull shape parameter (k) derived from statistical distribution fitting

Mechanism: Wind power extraction is non-linearly proportional to the cube of wind speed. Coarsening temporal resolution applies temporal averaging, which compresses the variance of the observed wind speed distribution. Because the expectation of the cube exceeds the cube of the expectation (E[v³] > (E[v])³), averaging reduces the estimated mean power. This variance compression simultaneously shifts the fitted Weibull shape parameter downward, which propagates through the standard wind power density integral. Longer time series partially compensate by capturing more extreme events, but cannot correct the bias introduced by the averaging operator itself.

grounded on: R704955, R707531, R707519, R707596, R707568, R708270, R709006, R708940

Integrating dew point depression and cloudiness index into a boosting-based ensemble model significantly reduces global solar radiation prediction errors during high-variability seasons (spring and autumn) compared to nonlinear autoregressive time-series models.

IV: Model feature engineering and architecture (Ensemble boosting with thermodynamic/cloud indices vs. Nonlinear autoregressive time-series with historical radiation only) · DV: Mean Absolute Error (MAE) of global solar radiation forecasts

Mechanism: Spring and autumn exhibit high meteorological variability due to rapid transitions in atmospheric moisture and cloud cover, which historical irradiance time-series models fail to capture due to their reliance on temporal inertia. Dew point depression (derived from temperature and dew point) serves as a proxy for atmospheric instability and convective potential, while cloudiness index directly quantifies radiative遮挡. An ensemble boosting framework iteratively reweights these physical predictors to correct the systematic biases of individual base learners, thereby enhancing adaptability to sudden atmospheric shifts that autoregressive models cannot anticipate.

grounded on: R1563739, R1563745, R1563736, R1563748, R1563694

The application of heterogeneous mixture distributions to model wind speed data reduces the uncertainty in estimated annual wind energy yield by 15-30% compared to single-parameter probability distributions, due to improved capture of multi-modal wind regimes.

IV: Distribution model type (Single-parameter vs. Heterogeneous mixture) · DV: Uncertainty in annual wind energy yield estimation (quantified as the relative width of the 95% confidence interval)

Mechanism: Single-parameter distributions assume unimodality and often misrepresent the tails and peaks of wind speed data in complex climates. Mixture distributions decompose the wind speed data into multiple components, accurately fitting each mode. This reduces the residual error in the probability density function estimation. When propagated through the power curve to calculate energy yield, the reduced PDF error narrows the statistical bounds of the AEP estimate, thereby lowering uncertainty.

grounded on: R709096, R707542, R707692, R705610