We introduce Support Vector Generation (SVG), a kernel-based framework that converts a frozen language model into an interpretable, training-free classifier for zero- and few-shot learning. SVG operates by combining Metropolis–Hastings sampling with support vector machine optimization in the reproducing kernel Hilbert space (RKHS) induced by the language model's embedding. Each classification decision is based on a weighted combination of at most 32 natural-language sentences, which serve as explicit support vectors and provide faithful rationales. Our theoretical analysis proves that SVG minimizes the empirical hinge loss over the span of the supports and admits a generalization bound independent of the language model size. Experiments on the GLUE benchmark show that SVG matches or surpasses prompting-based zero-shot baselines in accuracy across multiple tasks—without any fine-tuning or GPU acceleration. Notably, our CPU-only implementation completes training in under three minutes per task, and maintains competitive inference speed. These results suggest that SVG offers a viable path toward efficient, interpretable NLP systems under compute constraints.

Our study aims to strengthen truthfulness of the two-path mechanism: an information diffusion algorithm to find an influential node in non-cooperative directed acrylic graphs (DAGs). This subject is important because the two-path mechanism ensures only weak truthfulness (i.e., nodes are indifferent between reporting true or false out-edges), which restricts node selection accuracy. To enhance the mechanism, we employed an additional reward layer based on a multi-task peer prediction, where an informative equilibrium provides strictly higher rewards than any other equilibrium in virtually all cases (strong truthfulness). Rewards, which are derived from a comparison of each report, encourage a node to report true out-edges without affecting its own probability of being selected by the original two-path mechanism. We have also experimentally confirmed that our proposed strongly truthful two-path mechanism can sufficiently elicit true out-edges from each node.

When querying RDF and RDFS data, for improving the performance, it is common to derive all triples according to RDFS entailment rules before query processing. An undesirable drawback of this approach is that a large number of triples are generated by the RDFS reasoning, and hence considerable amount of storage space is required if we materialize the RDFS closure. In this paper, we propose RDF packages, which is a time and space efficient format for RDF data. In an RDF package, a set of triples of the same class or triples having the same predicate are grouped into a dedicated node named Package. Using Packages, we can represent any metadata that can be expressed by RDF. An important feature of the RDF packages is that, when performing RDFS reasoning, the same rules can be applied without any modification, thereby allowing us to use existing RDFS reasoners. In this paper, we discuss the model of RDF packages and its rules, followed by the transformation between RDF and RDF packages. We also discuss the implementation RDF packages using an existing RDF framework. Finally, we demonstrate the performance of the proposed scheme in triple size, reasoning speed, and querying speed.