Searcharxiv⌕ Search

arXiv subjects

Sina Barazandeh

Publications and source records attributed to Sina Barazandeh.

4 recordsLinked to original sources

Late but Valuable: Why Deadlines Change the Measured Value of LLM Reasoning

Large language models (LLMs) increasingly plan and run experiments in automated laboratories. These experiments keep running while the model reasons and run on schedules with hard deadlines. If an LLM-generated answer arrives late, the system executes a safe fallback action instead, such as rejecting a sample. This creates a systems problem that standard serving metrics do not capture. An answer that matches the fallback adds no value, and answers that differ from the fallback may require more reasoning time. Reporting answer quality and latency separately can therefore overstate the value of reasoning. We test this on synthetic sample decisions, where the best choice is known exactly and a runtime enforces the deadline. We compare one LLM's fast mode, which answers directly, with its thinking mode, which reasons first. In a prespecified replay of 1,200 new decisions, thinking answers that missed a deadline were worth about five times as much as on-time ones. Multiplying the on-time rate by the mean answer value predicted about twice the value that thinking delivered. This prediction overstated thinking's advantage over fast mode by 0.0241 per decision, and by 0.0191 in a repeat on 300 new decisions. Both 95% confidence intervals excluded zero. In a single-GPU Qwen3-8B experiment using simultaneous request bursts, raising the number of requests from 8 to 56 reduced the share of thinking answers that met a 45 s deadline from 79.2% to 1.4%, although total output tripled. Evaluations of LLM agents for laboratories should therefore measure each answer's value and arrival time together and score the action that actually runs.

cs.LG↗

ShiftCache: Fast and Exact Indel Effect Prediction by Reusing Shifted Reference Encodings

Sequence-to-function models such as AlphaGenome and Borzoi read up to a megabase of DNA and score a genetic variant by comparing their predictions for the reference and alternate sequences. Insertions and deletions (indels) are costly to score, because an indel usually shifts the unchanged DNA after it onto a different pooling grid, which prevents direct reuse of the reference's encoder activations. The AlphaGenome Atlas scored more than 100 million indels with three full model passes each, although the DNA after an indel matches the reference shifted by the indel's length. We present ShiftCache, an exact reuse system that encodes each shifted reference once and shares it among all variants scored in the same window with the same shift. ShiftCache copies every encoder output a variant can reuse and recomputes the rest in batched tiles. On held-out AlphaGenome windows shared by many indels, in strict fp32 on one A6000, ShiftCache is 3.5 to 6.5$\times$ faster than deduplicated dense execution and 1.6 to 2.7$\times$ faster than our adaptation of layer-by-layer reuse to indels. It is 3.5 to 4.1$\times$ faster than dense reimplementations of two published analyses with Borzoi and AlphaGenome, and 2.6$\times$ faster than dense execution with Enformer. Scoring each variant in its own centered 131 kb window makes ShiftCache only 1.36$\times$ faster than deduplicated dense execution, and highly diverse insertion lengths or many dispersed edits can make it slower than layer-by-layer reuse or dense execution. For encoders that meet our conditions, ShiftCache computes the same predictions as dense execution in exact arithmetic. On an A6000, every checked output of the default AlphaGenome and Borzoi runs matches dense execution within an absolute and relative tolerance of $10^{-5}$. With larger tile batches, with Enformer or on an H100, a small fraction of values exceeds this tolerance.

cs.LG↗

Deep Learning meets Blockchain for Automated and Secure Access Control

Access control is a critical component of computer security, governing access to system resources. However, designing policies and roles in traditional access control can be challenging and difficult to maintain in dynamic and complex systems, which is particularly problematic for organizations with numerous resources. Furthermore, traditional methods suffer from issues such as third-party involvement, inefficiency, and privacy gaps, making transparent and dynamic access control an ongoing research problem. Moreover detecting malicious activities and identifying users who are not behaving appropriately can present notable difficulties. To address these challenges, we propose DLACB, a Deep Learning Based Access Control Using Blockchain, as a solution to decentralized access control. DLACB uses blockchain to provide transparency, traceability, and reliability in various domains such as medicine, finance, and government while taking advantage of deep learning to not rely on predefined policies and eventually automate access control. With the integration of blockchain and deep learning for access control, DLACB can provide a general framework applicable to various domains, enabling transparent and reliable logging of all transactions. As all data is recorded on the blockchain, we have the capability to identify malicious activities. We store a list of malicious activities in the storage system and employ a verification algorithm to cross-reference it with the blockchain. We conduct measurements and comparisons of the smart contract processing time for the deployed access control system in contrast to traditional access control methods, determining the time overhead involved. The processing time of DLBAC demonstrates remarkable stability when exposed to increased request volumes.

cs.CR↗

DLACB: Deep Learning Based Access Control Using Blockchain

In general, deep learning models use to make informed decisions immensely. Developed models are mainly based on centralized servers, which face several issues, including transparency, traceability, reliability, security, and privacy. In this research, we identify a research gap in a distributed nature-based access control that can solve those issues. The innovative technology blockchain could fill this gap and provide a robust solution. Blockchain's immutable and distributed nature designs a useful framework in various domains such as medicine, finance, and government, which can also provide access control as opposed to centralized methods that rely on trusted third parties to access the resources. In existing frameworks, a traditional access control approach is developed using blockchain, which depends on predefined policies and permissions that are not reliable. In this research, we propose DLACB: Deep Learning Based Access Control Using Blockchain, which utilizes a deep learning access control mechanism to determine a user's permissions on a given resource. This proposed framework authenticates the users and logs the access requests on the blockchain to recognize malicious users. The results show that this proposed framework operates correctly for all possible scenarios.

cs.CR↗