This paper presents Retrieval-Augmented Forecasting (RAF), a novel framework for tabular time-series prediction that dynamically retrieves and integrates relevant historical table slices. RAF addresses three key limitations of existing methods: 1) schema rigidity through dynamic hashing of column metadata, 2) temporal myopia via cross-attention with learned decay, and 3) pipeline sub-optimality via end-to-end retriever-forecaster co-training. Experiments across macroeconomic (FRED-MD), financial (Yahoo Finance), and development (WorldBank) benchmarks demonstrate RAF’s superiority over six baselines, reducing sMAPE by 19.1-26.5% while maintaining robustness to schema changes (+3.2% sMAPE increase vs. +6.7-12.7% for alternatives). The architecture’s computational overhead (1.8 vs. 1.2 hours/epoch vs. TFT) is justified by significant accuracy gains in critical scenarios like market shocks (61.7% vs. 55.1% directional accuracy).

Mathematical information retrieval requires understanding the complex relationship between natural language and formulae. This paper presents a benchmarking study on Formula-Text Cross-Retrieval, comparing a sparse baseline (BM25), off-the-shelf dense embeddings (OpenAI, BGE), and a fine-tuned dual-encoder model. Our model, trained with a contrastive objective on the ARQAR dataset, significantly outperforms all baselines, achieving state-of-the-art results. Ablation studies confirm the importance of linearization, a shared-weight architecture, and the Multiple Negatives Ranking loss. The work provides a strong foundation for mathematical NLP applications.

This paper presents a knowledge-grounded framework for cryptocurrency scam detection using retrieval-augmented language models. We address three key limitations of existing approaches: static knowledge bases, unreliable LM outputs, and fixed classification thresholds. Our method combines (1) temporally-weighted retrieval from scam databases, (2) confidence-aware fusion of parametric and external knowledge, and (3) adaptive threshold optimization via gradient ascent. Experiments on CryptoScams and Twitter Financial Scams datasets demonstrate state-of-the-art performance, with 22% higher recall at equivalent precision compared to fixed thresholds, 4.3× lower hallucination rates than pure LMs, and 89% temporal performance retention on emerging scam types. The system achieves real-time operation (45ms/query) while maintaining interpretability through evidence grounding. Ablation studies confirm each component’s necessity, with confidence fusion proving most critical (12.1% performance drop when removed). These advances enable more robust monitoring of evolving cryptocurrency threats while addressing fundamental challenges in knowledgeable foundation models.