{"id":"arxiv/2609.25778","url":"https://arxiv.org/abs/2609.25778","tags":["stat.ML","cs.LG","math.ST"],"score":0,"title":"Statistical Gains from Looped Estimation under Parameter Budgets","domain":"stat.ML","source":"arxiv","authors":["Xinyu Tian","Xiaotong Shen"],"reviews":0,"subject":"stat.ML","summary":"Memory constraints in artificial intelligence motivate accurate function approximation with fewer parameters. We study looping, which repeatedly composes one update function with shared parameters; each output becomes the next input. A looped Transformer, for example, reuses one block, whereas its conventional untied counterpart uses separately parameterized blocks. We compare their parameter requirements for a given worst-case approximation accuracy, or equivalently, their approximation accuracy under a common budget limiting distinct trainable coefficients. We then ask whether this…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-09-22","reactions":{}}