arXiv subjects

Yuliana Shakhvalieva

Publications and source records attributed to Yuliana Shakhvalieva.

3 recordsLinked to original sources

arXiv2026

What Limits Recursive Reasoning Models: Optimization, Architecture and Test-Time Scaling

Recursive reasoning models apply a small shared Transformer block many times to refine a latent state. This gives them large effective depth with few parameters and makes them strong on algorithmic tasks. Such compact solvers are natural candidates for tools that an LLM can call on narrow algorithmic subproblems. However, existing models such as HRM, TRM and URM differ in architecture, gradient propagation and training procedure simultaneously. This makes it hard to tell what drives their performance, and their optimization is still poorly understood and often unstable. In this work we address both of these gaps. First, we study these questions under a unified experimental pipeline spanning six algorithmic domains. Individual controlled ablations are performed on representative domains, while the resulting recipe is evaluated across the full suite. The study reveals a surprisingly simple recipe for stable and generalizable recursive reasoning: an intermediate gradient horizon, large physical batches and controlled updates of the recurrent state. An explicit hierarchical architecture is not needed. Second, we combine these findings into a stable 13.6M-parameter model that achieves the strongest overall performance among the evaluated recursive baselines, with particularly large gains on out-of-distribution generalization. It raises Arithmetic OOD accuracy to 71.2%, from 36.2% for the strongest baseline, while reaching 98.41% on Sudoku and 59.5% pass@2 on ARC-AGI-1. Our results show that, within the recursive architectures studied here, performance depends strongly on how recurrence is optimized and stabilized. More broadly, it shows how AI systems can be improved by optimizing their components one at a time.

cs.LG↗

arXiv2026

Closing the Loop: Practical Training Recipes for Looped Language Models

Looped language models increase effective depth by repeatedly applying a shared block of layers, but existing large-scale recipes require multi-stage training over trillions of tokens, while the benefits of recurrence remain difficult to separate from differences in data and training. In this work, we establish practical training recipes for looped language models, with three main results. (1) We develop a compute-efficient from-scratch pipeline that reduces the training budget from 7.7T tokens in Ouro to 310B tokens while retaining strong reasoning performance. Pretraining followed by high-quality mid-training, together with learning-rate warmup and stronger exit-gate regularization, enables stable recurrent training without prior multi-stage schedules. (2) Under controlled comparisons, our 1.4B LoopLM outperforms a parameter-matched dense model trained on the same data and token budget on all 12 evaluated benchmarks, including +14 points on GSM8K, +10 on MATH, and +22 on DROP. At matched inference compute, it approaches a 3.9B dense model on mathematical reasoning and reading comprehension while using only 36\% as many parameters. (3) We introduce a minimal recipe for converting pretrained dense models into looped ones: a single learned input-mixing scalar and a smoothed exit loss, with no step-specific parameters. Applied to Qwen3-1.7B-Base, Looped Qwen improves over an identically continued dense baseline on every evaluated benchmark across two data regimes, with statistically clear gains on GSM8K, MATH, and MMLU-Pro on the curated mixture. Together, these results make looped language models substantially cheaper to train from scratch and practical to introduce into existing pretrained checkpoints, while isolating the gains due to recurrence itself.

cs.CL↗

arXiv2026

Financial Transaction Retrieval and Contextual Evidence for Knowledge-Grounded Reasoning

Nowadays, success of financial organizations heavily depends on their ability to process digital traces generated by their clients, e.g., transaction histories, gathered from various sources to improve user modeling pipelines. As general-purpose LLMs struggle with time-distributed tabular data, production stacks still depend on specialized tabular and sequence models with limited transferability and need for labeled data. To address this, we introduce FinTRACE, a retrieval-first architecture that converts raw transactions into reusable feature representations, applies rule-based detectors, and stores the resulting signals in a behavioral knowledge base with graded associations to the objectives of downstream tasks. Across public and industrial benchmarks, FinTRACE substantially improves low-supervision transaction analytics, doubling zero-shot MCC on churn prediction performance from 0.19 to 0.38 and improving 16-shot MCC from 0.25 to 0.40. We further use FinTRACE to ground LLMs via instruction tuning on retrieved behavioral patterns, achieving state-of-the-art LLM results on transaction analytics problems.

cs.IR↗