Skip to main content
Search

# Online Learning

7 related articles

Subscribe to updates
  1. arXiv

    Delayed feedback meets a memory cap: CA-OCRC keeps online risk control on target with capacity growing only logarithmically in the maximum delay

    Online conformal risk control has assumed each round's loss arrives immediately and intact, yet real systems delay feedback and can track only a limited number of pending predictions, so some feedback is lost for good. CA-OCRC designs the tracking policy and the prediction sets together: each round draws a random proxy delay, is tracked while capacity allows until its feedback arrives or the proxy delay expires, and a two-level inverse probability weighting compensates for lost feedback. Risk bounds for arbitrary and stochastic delays follow, and optimizing the proxy delay distribution shows that capacity growing only logarithmically in the maximum delay suffices to match the convergence rate of OCRC with unlimited feedback.

  2. arXiv

    Debate as a training protocol gets its first learning analysis: cooperating debaters can cut mistakes to a constant in some cases

    When debate is used as a training protocol, a human judges which of two debaters is right instead of supplying labels. Earlier work analyzed only the game between debaters under a known judge, leaving open how debaters learn the correct function. This work formulates debate as an online learning problem and shows there are cases where adversarial debate cannot learn but cooperative debate needs only a constant number of mistakes, with a new combinatorial dimension characterizing the optimal mistake bound for cooperative debaters in the realizable setting.

  3. Coastal buoys get a three-level wave-height quality flag: one temporal model transfers across stations at a mean binary F1 of 0.856

    Significant wave-height records from a Brazilian coastal buoy network can now receive an hourly three-level quality flag, GOOD, SUSPECT or BAD, instead of relying only on fixed physical-limit and rate-of-change thresholds. The model decides one-sidedly from the current value and the preceding 168 hours, reaching a binary F1 of 0.853 on a pooled controlled-failure test set across six buoys and a mean binary F1 of 0.856 when transferred to unseen stations. In hourly streaming simulation, abrupt, multiplicative and sentinel failures were detected at onset, while drift and stuck-sensor episodes required accumulated evidence.

  4. arXiv

    AFA-Bandit casts online active feature acquisition as a combinatorial bandits-with-knapsacks problem, and LP-Chain beats HEDGE-based BwK and a deep RL baseline on synthetic data with linear-size chain search

    The work formulates online active feature acquisition (AFA) as a predictor-coupled combinatorial Bandits with Knapsacks problem that jointly decides feature acquisition and label prediction under a global budget, and derives a tighter regret upper bound than standard BwK by exploiting a cardinality-aware confidence bound and subset feedback; to avoid an exponentially large action space, the authors propose LP-Chain, which searches a cost-aware chain of feature subsets whose size grows linearly with the number of features, and on synthetic Gaussian-mixture data it attains lower prediction error than HEDGE-based BwK and a deep RL online AFA baseline.

  5. arXiv

    Systematic evaluation of streaming tabular foundation models: TabFM+FIFO leads on nearly all nine streams but serves over an order of magnitude slower than the fastest ensemble

    In a unified prequential (test-then-train) protocol across real and synthetic streams, this study systematically compares tabular foundation models (TFMs) with streaming learners and streaming AutoML, finding that TFMs reach the highest predictive accuracy, recover faster after drift, and stay most accurate under label delay, but cost more than an order of magnitude more to serve; memory size and backbone quality matter more than the memory management policy, and repeated context encoding is the dominant cost, which a compact causal model with cached encoded context can hold nearly constant across memory sizes at close accuracy.

  6. arXiv

    JOVE jointly assigns executor LLMs and paid verification, cutting average cost and latency by at least 3.17x while staying competitive on accuracy across four reasoning benchmarks

    JOVE is an online framework that decomposes complex reasoning queries into directed acyclic task graphs and distributes them across heterogeneous LLMs while jointly deciding which intermediate outputs to send for paid verification; it makes execution and verification decisions by solving a sequence of per-query mixed-integer linear programs, updates task-dependent LLM quality estimates online from verification feedback, and adds an information-gain bonus that folds the value of learning into allocation, achieving sublinear quality-learning regret under a long-term budget and a per-query latency constraint, with competitive accuracy against standard inference baselines and at least 3.17x lower average cost and latency across four reasoning benchmarks.

  7. arXiv

    Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data

    The work proposes an "infinite-parameter LLM" architecture in which a compact hypernetwork encodes run-time data (facts, instructions, demonstrations) into a low-dimensional latent code, that code generates a low-rank additive modulation of a shared base feed-forward network so each token's "expert" is generated rather than drawn from a stored bank, and a Bayesian belief over the latent code is carried and updated online so the effective weight keeps evolving within a session; the authors also specify an evaluation protocol that pits carrying knowledge in the weights against carrying it in the prompt at matched budget.

You’re all caught up