# Distribution Shift

Clinical prediction models inherit their quality from routine-care training data, yet missingness rates and class ratios are not tied to what happens after deployment. An exploratory benchmark across three MIMIC-IV cohorts and the multi-institution eICU database splits data quality into completeness, label noise, distributional similarity, representativeness and temporal integrity, training a real model under each controlled condition and measuring held-out log-loss and calibration change. The composite moved in the expected direction in all three controlled cohorts, but combined corruptions were sub-additive rather than multiplicative, and across 21 analyzable eICU hospitals it tracked calibration but not detectably discrimination.

Clinical prediction models inherit their quality from routine-care training data, yet missingness rates and class ratios are not tied to what happens after deployment. An exploratory benchmark across three MIMIC-IV cohorts and the multi-institution eICU database splits data quality into completeness, label noise, distributional similarity, representativeness and temporal integrity, training a real model under each controlled condition and measuring held-out log-loss and calibration change. The composite moved in the expected direction in all three controlled cohorts, but combined corruptions were sub-additive rather than multiplicative, and across 21 analyzable eICU hospitals it tracked calibration but not detectably discrimination.

Memento sends sensor observations by reference: the device searches a vector table in persistent memory for a similar past sample and, on a hit, transmits only an index, and modeling shows that on low-rate links such as LoRa a 25% hit rate cuts latency and energy by roughly a quarter.

Memento sends sensor observations by reference: the device searches a vector table in persistent memory for a similar past sample and, on a hit, transmits only an index, and modeling shows that on low-rate links such as LoRa a 25% hit rate cuts latency and energy by roughly a quarter.

In long-horizon autoregressive reinforcement learning, local policy mismatch compounds along the trajectory into state distribution shift, destabilizing training and eventually collapsing it as the horizon grows. Trajectory-Aware Policy Optimization derives a state-distribution correction from the off-policy policy gradient and writes it into the objective, estimates it with a geometric mean to cut variance, and tightens the clipping boundary with the length of the preceding trajectory. Across eight competition-level math and code benchmarks it improves over GRPO by up to 48% on math and 12% on code on average, with training scores rising steadily while KL divergence stays controlled.

In long-horizon autoregressive reinforcement learning, local policy mismatch compounds along the trajectory into state distribution shift, destabilizing training and eventually collapsing it as the horizon grows. Trajectory-Aware Policy Optimization derives a state-distribution correction from the off-policy policy gradient and writes it into the objective, estimates it with a geometric mean to cut variance, and tightens the clipping boundary with the length of the preceding trajectory. Across eight competition-level math and code benchmarks it improves over GRPO by up to 48% on math and 12% on code on average, with training scores rising steadily while KL divergence stays controlled.

On strictly paired vulnerable and patched C/C++ functions, whitened CodeBERT layer-3 embeddings with nearest-neighbour distance separate unfamiliar samples from familiar ones at a mean OOD AUROC of 0.739, and expose a cross-project patch false-positive rate as high as 0.507.

On strictly paired vulnerable and patched C/C++ functions, whitened CodeBERT layer-3 embeddings with nearest-neighbour distance separate unfamiliar samples from familiar ones at a mean OOD AUROC of 0.739, and expose a cross-project patch false-positive rate as high as 0.507.

The work reads pretrained-weight spectral structure through 15 (architecture x pretraining-strategy) cells, proves that the OOD accuracy gap is bounded by how tightly source representations concentrate, and shows that a single statistic selected per cell orders OOD robustness across 116 models and 7 modalities, with the cell, metric and sign locked before evaluation for out-of-matrix families including EEG, genomic and protein.

The work reads pretrained-weight spectral structure through 15 (architecture x pretraining-strategy) cells, proves that the OOD accuracy gap is bounded by how tightly source representations concentrate, and shows that a single statistic selected per cell orders OOD robustness across 116 models and 7 modalities, with the cell, metric and sign locked before evaluation for out-of-matrix families including EEG, genomic and protein.

The work proposes a model-agnostic online martingale PAC-Bayesian framework that replaces independent-sample concentration with martingale concentration to yield finite-sample certificates under temporal dependence and distribution shift, and uses the certificate as a surrogate regularizer for online calibration by training a gated residual Bayesian head on top of a fixed forecasting backbone under a predict-then-update protocol, improving stability and accuracy under covariate and concept shift across convolutional, attention-based, and large language model-based forecasters.

The work proposes a model-agnostic online martingale PAC-Bayesian framework that replaces independent-sample concentration with martingale concentration to yield finite-sample certificates under temporal dependence and distribution shift, and uses the certificate as a surrogate regularizer for online calibration by training a gated residual Bayesian head on top of a fixed forecasting backbone under a predict-then-update protocol, improving stability and accuracy under covariate and concept shift across convolutional, attention-based, and large language model-based forecasters.

The work sizes the selection cost of picking the best single model as comparator on the evaluation data in safety-routing benchmarks: 0.003-0.030 of harm under random splits and 0.045-0.113 under held-out categories on HELM Safety, comparable to the whole deficit attributed to routing, rising seven- to ninefold on AgentDojo when suites are held out; scored honestly under shift, routing buys little, and an attacker who knows which model it faces lowers GPT-5.4's judged recognition by 19.6 points.

The work sizes the selection cost of picking the best single model as comparator on the evaluation data in safety-routing benchmarks: 0.003-0.030 of harm under random splits and 0.045-0.113 under held-out categories on HELM Safety, comparable to the whole deficit attributed to routing, rising seven- to ninefold on AgentDojo when suites are held out; scored honestly under shift, routing buys little, and an attacker who knows which model it faces lowers GPT-5.4's judged recognition by 19.6 points.