Centre for Language and Semantic Technologies

Publications

• Book

  13th International Conference, AIST 2026, Astana, Kazakhstan, October 16–18, 2026, Proceedings. Analysis of Images, Social Networks and Texts. AIST 2026. Lecture Notes in Computer Science (LNCS, volume 16795)

  Vol. 16795. Springer, 2026.

• Article

  Zhang Z., Ilvovsky D.

  PrimeKV: Preserving critical KV cache without attention scores for long-context inference

  Efficient management of the Key-Value (KV) cache is essential for deploying large language models (LLMs) in long-context scenarios, where memory requirements often exceed the capacity of resource-constrained hardware. Many existing training-free compression algorithms typically rely on attention-driven heuristics, but exhibit two key limitations. First, they require explicit attention score computation, which is difficult to reconcile with optimized inference kernels that avoid materializing the full attention matrix. Second, they depend on trailing observation windows as anchors, leading to degraded performance in task-agnostic settings where future queries remain unpredictable. To address these issues, we propose PrimeKV, a parameter-free KV cache compression algorithm that identifies critical KV pairs using a perturbation-inspired importance estimate. Because PrimeKV relies only on intrinsic KV cache features and does not require materialized attention scores, its scoring design is algorithmically compatible with attention-score-free inference kernels such as FlashAttention and PagedAttention, while validated system integration and serving efficiency require separate engineering and evaluation. PrimeKV also preserves model fidelity without relying on explicit query anchors. Extensive experiments on RULER, LongBench, multi-turn conversation tasks, and AIME25 demonstrate strong overall performance of PrimeKV against representative KV cache compression baselines. Compared with the strongest attention-based baseline, PrimeKV achieves up to a 244.6% relative improvement on RULER and 11.54% on LongBench, rising to 28.62% on multi-turn conversation tasks. On AIME25, PrimeKV achieves the same performance as Full KV even under KV cache compression.

  Neurocomputing. 2027. Vol. 710.

• Book chapter

  Zhang Z., Ilvovsky D.

  Thesis Proposal: Efficient KV Cache Reuse for Multi-Document Retrieval-Augmented Generation

  Retrieval-Augmented Generation (RAG) systems face efficiency bottlenecks in prefill due to attention mechanism, and traditional KV cache only accelerates decoding. In this context, reusing document-level KV cache computed for retrieved documents in previous sessions during the prefill stage appears to be a natural way to amortize computation, but it raises serious correctness challenges due to position and context misalignment across queries and sessions. This research proposes a multi-document KV cache reuse framework for multi-document RAG workloads across queries and sessions to resolve position misalignment and context misalignment, preserving accuracy while eliminating document-specific quadratic complexity in prefill. Theoretical analysis will establish conditions under which multi-document KV cache reuse remains stable and close to full recomputation, providing principled guarantees for both efficiency and accuracy. These results will enable deployment in existing RAG pipelines without architectural changes or model retraining. Crucially, to ensure robustness in real-world deployments, validation will extend beyond standard benchmarks to include noise-robustness tests and domain-specific workloads (e.g., legal). The research aims to empirically confirm these guarantees and demonstrate that substantial prefill speedups can be achieved without materially degrading task-level performance.

  In bk.: Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop). Vol. 4. Association for Computational Linguistics, 2026. P. 160-169.

• Working paper

  Меньшиков И. А., Бернадотт А. К., Elvimov N. S.

  Hessian-based lightweight neural network for brain vessel segmentation on a minimal training dataset

  Accurate segmentation of blood vessels in brain magnetic resonance angiography (MRA) is essential for successful surgical procedures, such as aneurysm repair or bypass surgery. Currently, annotation is primarily performed through manual segmentation or classical methods, such as the Frangi filter, which often lack sufficient accuracy. Neural networks have emerged as powerful tools for medical image segmentation, but their development depends on well-annotated training datasets. However, there is a notable lack of publicly available MRA datasets with detailed brain vessel annotations. To address this gap, we propose a novel semi-supervised learning lightweight neural network with Hessian matrices on board for 3D segmentation of complex structures such as tubular structures, which we named HessNet. The solution is a Hessian-based neural network with only 6000 parameters. HessNet can run on the CPU and significantly reduces the resource requirements for training neural networks. The accuracy of vessel segmentation on a minimal training dataset reaches state-of-the-art results. It helps us create a large, semi-manually annotated brain vessel dataset of brain MRA images based on the IXI dataset (annotated 200 images). Annotation was performed by three experts under the supervision of three neurovascular surgeons after applying HessNet. It provides high accuracy of vessel segmentation and allows experts to focus only on the most complex important cases. The dataset is available at https://git.scinalytics.com/terilat/VesselDatasetPartly.

  Statistical mechanics. arXie. arXive, 2025