{"slug": "a-classifier-that-teaches-itself-self-improving-frozen-gate-training-sift-for", "title": "A Classifier That Teaches Itself: Self-Improving, Frozen-gate Training (SIFT) for Dynamic Document Classification", "summary": "Researchers present SIFT (Self-Improving, Frozen-gate Training), a dynamic classifier service that uses a cheap CPU-bound pipeline with a SPLADE sparse encoder and LightGBM head, escalating low-confidence cases to an LLM judge whose verdicts expand the labeled corpus and improve accuracy over time. SIFT eliminates upfront labeling projects by requiring only a declarative bundle for new document families, and ensures safe autonomous retraining through a two-part promote gate that checks critical-label F1 regression and a frozen golden regression set. The system's marginal labeling cost trends toward zero as it continuously learns from production traffic.", "body_md": "arXiv:2607.18358v1 Announce Type: new\nAbstract: Document classification is a solved problem in the laboratory and an unsolved one in the enterprise. The blocker is rarely model architecture; it is the labeling project that must precede a model and the institutional fear of letting a model retrain itself once one exists. We present SIFT (Self-Improving, Frozen-gate Training), a dynamic classifier service, which attacks both. SIFT serves classification from a deliberately cheap, CPU-bound pipeline, a SPLADE sparse encoder feeding a LightGBM head, and escalates only the low-confidence minority of pages to an LLM judge. The judge's verdicts are written back into a labeled corpus, so the expensive model continuously teaches the cheap one: the escalation rate falls, the corpus grows from production traffic rather than from an up-front annotation effort, and accuracy compounds with use. Onboarding a new document family requires only a declarative bundle, label space, anchor phrases, and a judge glossary, not a labeling project. The harder problem is safety: an autonomously retraining classifier can silently regress. SIFT resolves this with a two-part promote gate, a critical-label F1 regression check plus a frozen golden regression set the model is never trained on, either of which vetoes promotion. This turns \"retrain monthly without a human\" from reckless into routine. We describe the architecture, the self-feeding corpus loop, the frozen-gate promotion mechanism, and an illustrative multi-domain deployment, and we discuss the economics of a classifier whose marginal labeling cost trends toward zero.", "url": "https://wpnews.pro/news/a-classifier-that-teaches-itself-self-improving-frozen-gate-training-sift-for", "canonical_source": "https://www.machinebrief.com/news/a-classifier-that-teaches-itself-self-improving-frozen-gate-2pli", "published_at": "2026-07-22 04:00:00+00:00", "updated_at": "2026-07-22 04:08:57.033232+00:00", "lang": "en", "topics": ["machine-learning", "large-language-models", "ai-products", "ai-infrastructure"], "entities": ["SIFT", "SPLADE", "LightGBM", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/a-classifier-that-teaches-itself-self-improving-frozen-gate-training-sift-for", "markdown": "https://wpnews.pro/news/a-classifier-that-teaches-itself-self-improving-frozen-gate-training-sift-for.md", "text": "https://wpnews.pro/news/a-classifier-that-teaches-itself-self-improving-frozen-gate-training-sift-for.txt", "jsonld": "https://wpnews.pro/news/a-classifier-that-teaches-itself-self-improving-frozen-gate-training-sift-for.jsonld"}}