arXiv subjects

Sheng Cao

Publications and source records attributed to Sheng Cao.

12 recordsLinked to original sources

arXiv2026

Case-Based Calibration of Adaptive Reasoning and Execution for LLM Tool Use

Tool use extends large language models beyond parametric knowledge, but reliable execution requires balancing appropriate reasoning depth with strict structural validity. We approach this problem from a case-based perspective to present CAST, a case-driven framework that treats historical execution trajectories as structured cases. Instead of reusing raw exemplar outputs, CAST extracts case-derived signals to identify complexity profiles for estimating optimal reasoning strategies, alongside failure profiles to map likely structural breakdowns. The framework translates this knowledge into a fine-grained reward design and adaptive reasoning, enabling the model to autonomously internalize case-based strategies during reinforcement learning. Experiments on BFCLv2 and ToolBench demonstrate that CAST improves both schema-faithful execution and task-level tool-use success while reducing unnecessary deliberation. The approach achieves up to 5.85 percentage points gain in overall execution accuracy and reduces average reasoning length by 26%, significantly mitigating high-impact structural errors. Ultimately, this demonstrates how historical execution cases can provide reusable adaptation knowledge for calibrated tool use.

cs.AI↗

arXiv2026

The Auton Agentic AI Framework

The field of Artificial Intelligence is undergoing a transition from Generative AI -- probabilistic generation of text and images -- to Agentic AI, in which autonomous systems execute actions within external environments on behalf of users. This transition exposes a fundamental architectural mismatch: Large Language Models (LLMs) produce stochastic, unstructured outputs, whereas the backend infrastructure they must control -- databases, APIs, cloud services -- requires deterministic, schema-conformant inputs. The present paper describes the Auton Agentic AI Framework, a principled architecture for standardizing the creation, execution, and governance of autonomous agent systems. The framework is organized around a strict separation between the Cognitive Blueprint, a declarative, language-agnostic specification of agent identity and capabilities, and the Runtime Engine, the platform-specific execution substrate that instantiates and runs the agent. This separation enables cross-language portability, formal auditability, and modular tool integration via the Model Context Protocol (MCP). The paper formalizes the agent execution model as an augmented Partially Observable Markov Decision Process (POMDP) with a latent reasoning space, introduces a hierarchical memory consolidation architecture inspired by biological episodic memory systems, defines a constraint manifold formalism for safety enforcement via policy projection rather than post-hoc filtering, presents a three-level self-evolution framework spanning in-context adaptation through reinforcement learning, and describes runtime optimizations -- including parallel graph execution, speculative inference, and dynamic context pruning -- that reduce end-to-end latency for multi-step agent workflows.

cs.AI↗

arXiv2026

STEM: Scaling Transformers with Embedding Modules

Fine-grained sparsity promises higher parametric capacity without proportional per-token compute, but often suffers from training instability, load balancing, and communication overhead. We introduce STEM (Scaling Transformers with Embedding Modules), a static, token-indexed approach that replaces the FFN up-projection with a layer-local embedding lookup while keeping the gate and down-projection dense. This removes runtime routing, enables CPU offload with asynchronous prefetch, and decouples capacity from both per-token FLOPs and cross-device communication. Empirically, STEM trains stably despite extreme sparsity. It improves downstream performance over dense baselines while reducing per-token FLOPs and parameter accesses (eliminating roughly one-third of FFN parameters). STEM learns embedding spaces with large angular spread which enhances its knowledge storage capacity. More interestingly, this enhanced knowledge capacity comes with better interpretability. The token-indexed nature of STEM embeddings allows simple ways to perform knowledge editing and knowledge injection in an interpretable manner without any intervention in the input text or additional computation. In addition, STEM strengthens long-context performance: as sequence length grows, more distinct parameters are activated, yielding practical test-time capacity scaling. Across 350M and 1B model scales, STEM delivers up to ~3--4% accuracy improvements overall, with notable gains on knowledge and reasoning-heavy benchmarks (ARC-Challenge, OpenBookQA, GSM8K, MMLU). Overall, STEM is an effective way of scaling parametric memory while providing better interpretability, better training stability and improved efficiency.

cs.LG↗

arXiv2025

MobileLLM-R1: Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes

The paradigm shift in large language models (LLMs) from instinctive responses to chain-of-thought (CoT) reasoning has fueled two prevailing assumptions: (1) reasoning capabilities only emerge in sufficiently large models, and (2) such capabilities require training on massive datasets. While the first assumption has already been challenged by recent sub-billion-parameter reasoning models such as Qwen3-0.6B and DeepSeek distilled variants, the second remains largely unquestioned. In this work, we revisit the necessity of scaling to extremely large corpora (>10T tokens) for reasoning emergence. By carefully curating and resampling open-source datasets that we identify as beneficial under our designed metrics, we demonstrate that strong reasoning abilities can emerge with far less data. Specifically, we show that only ~2T tokens of high-quality data are sufficient, and pre-training with 4.2T tokens on the dataset resampled from these ~2T tokens, followed by a established post-training procedure, enables the development of MobileLLM-R1, a series of sub-billion-parameter reasoning models that substantially outperform prior models trained on fully open-sourced data. For example, MobileLLM-R1-950M achieves an AIME score of 15.5, compared to just 0.6 for OLMo-2-1.48B and 0.3 for SmolLM-2-1.7B. Remarkably, despite being trained on only 11.7% of the tokens compared to Qwen3's proprietary 36T-token corpus for pretraining, MobileLLM-R1-950M matches or surpasses Qwen3-0.6B across multiple reasoning benchmarks. To facilitate further research in this direction, we have made the models (https://huggingface.co/collections/facebook/mobilellm-r1) and code (https://github.com/facebookresearch/MobileLLM-R1) publicly available, along with the complete training recipe, data sources, and data mixing ratios.

cs.CL↗

arXiv2025

Automating Function-Level TARA for Automotive Full-Lifecycle Security

As modern vehicles evolve into intelligent and connected systems, their growing complexity introduces significant cybersecurity risks. Threat Analysis and Risk Assessment (TARA) has therefore become essential for managing these risks under mandatory regulations. However, existing TARA automation methods rely on static threat libraries, limiting their utility in the detailed, function-level analyses demanded by industry. This paper introduces DefenseWeaver, the first system that automates function-level TARA using component-specific details and large language models (LLMs). DefenseWeaver dynamically generates attack trees and risk evaluations from system configurations described in an extended OpenXSAM++ format, then employs a multi-agent framework to coordinate specialized LLM roles for more robust analysis. To further adapt to evolving threats and diverse standards, DefenseWeaver incorporates Low-Rank Adaptation (LoRA) fine-tuning and Retrieval-Augmented Generation (RAG) with expert-curated TARA reports. We validated DefenseWeaver through deployment in four automotive security projects, where it identified 11 critical attack paths, verified through penetration testing, and subsequently reported and remediated by the relevant automakers and suppliers. Additionally, DefenseWeaver demonstrated cross-domain adaptability, successfully applying to unmanned aerial vehicles (UAVs) and marine navigation systems. In comparison to human experts, DefenseWeaver outperformed manual attack tree generation across six assessment scenarios. Integrated into commercial cybersecurity platforms such as UAES and Xiaomi, DefenseWeaver has generated over 8,200 attack trees. These results highlight its ability to significantly reduce processing time, and its scalability and transformative impact on cybersecurity across industries.

cs.CR↗

arXiv2025

Param$\Delta$ for Direct Weight Mixing: Post-Train Large Language Model at Zero Cost

The post-training phase of large language models is essential for enhancing capabilities such as instruction-following, reasoning, and alignment with human preferences. However, it demands extensive high-quality data and poses risks like overfitting, alongside significant computational costs due to repeated post-training and evaluation after each base model update. This paper introduces $Param\Delta$, a novel method that streamlines post-training by transferring knowledge from an existing post-trained model to a newly updated base model with ZERO additional training. By computing the difference between post-trained model weights ($\Theta_\text{post}$) and base model weights ($\Theta_\text{base}$), and adding this to the updated base model ($\Theta'_\text{base}$), we define $Param\Delta$ Model as: $\Theta_{\text{Param}\Delta} = \Theta_\text{post} - \Theta_\text{base} + \Theta'_\text{base}$. This approach surprisingly equips the new base model with post-trained capabilities, achieving performance comparable to direct post-training. We did analysis on LLama3, Llama3.1, Qwen, and DeepSeek-distilled models. Results indicate $Param\Delta$ Model effectively replicates traditional post-training. For example, the $Param\Delta$ Model obtained from 70B Llama3-inst, Llama3-base, Llama3.1-base models attains approximately 95\% of Llama3.1-inst model's performance on average. $Param\Delta$ brings a new perspective on how to fully leverage models in the open-weight community, where checkpoints for base and instruct models are readily available and frequently updated, by providing a cost-free framework to accelerate the iterative cycle of model development.

cs.CL↗

arXiv2024

LLM-Augmented Retrieval: Enhancing Retrieval Models Through Language Models and Doc-Level Embedding

Recently embedding-based retrieval or dense retrieval have shown state of the art results, compared with traditional sparse or bag-of-words based approaches. This paper introduces a model-agnostic doc-level embedding framework through large language model (LLM) augmentation. In addition, it also improves some important components in the retrieval model training process, such as negative sampling, loss function, etc. By implementing this LLM-augmented retrieval framework, we have been able to significantly improve the effectiveness of widely-used retriever models such as Bi-encoders (Contriever, DRAGON) and late-interaction models (ColBERTv2), thereby achieving state-of-the-art results on LoTTE datasets and BEIR datasets.

cs.IR↗

arXiv2023

Multi-dimensional Data Quick Query for Blockchain-based Federated Learning

Due to the drawbacks of Federated Learning (FL) such as vulnerability of a single central server, centralized federated learning is shifting to decentralized federated learning, a paradigm which takes the advantages of blockchain. A key enabler for adoption of blockchain-based federated learning is how to select suitable participants to train models collaboratively. Selecting participants by storing and querying the metadata of data owners on blockchain could ensure the reliability of selected data owners, which is helpful to obtain high-quality models in FL. However, querying multi-dimensional metadata on blockchain needs to traverse every transaction in each block, making the query time-consuming. An efficient query method for multi-dimensional metadata in the blockchain for selecting participants in FL is absent and challenging. In this paper, we propose a novel data structure to improve the query efficiency within each block named MerkleRB-Tree. In detail, we leverage Minimal Bounding Rectangle(MBR) and bloom-filters for the query process of multi-dimensional continuous-valued attributes and discrete-valued attributes respectively. Furthermore, we migrate the idea of the skip list along with an MBR and a bloom filter at the head of each block to enhance the query efficiency for inter-blocks. The performance analysis and extensive evaluation results on the benchmark dataset demonstrate the superiority of our method in blockchain-based FL.

cs.DS↗

arXiv2023

Federated Learning Incentive Mechanism under Buyers' Auction Market

Auction-based Federated Learning (AFL) enables open collaboration among self-interested data consumers and data owners. Existing AFL approaches are commonly under the assumption of sellers' market in that the service clients as sellers are treated as scarce resources so that the aggregation servers as buyers need to compete the bids. Yet, as the technology progresses, an increasing number of qualified clients are now capable of performing federated learning tasks, leading to shift from sellers' market to a buyers' market. In this paper, we shift the angle by adapting the procurement auction framework, aiming to explain the pricing behavior under buyers' market. Our modeling starts with basic setting under complete information, then move further to the scenario where sellers' information are not fully observable. In order to select clients with high reliability and data quality, and to prevent from external attacks, we utilize a blockchain-based reputation mechanism. The experimental results validate the effectiveness of our approach.

cs.LG↗

arXiv2023

Portfolio-Based Incentive Mechanism Design for Cross-Device Federated Learning

In recent years, there has been a significant increase in attention towards designing incentive mechanisms for federated learning (FL). Tremendous existing studies attempt to design the solutions using various approaches (e.g., game theory, reinforcement learning) under different settings. Yet the design of incentive mechanism could be significantly biased in that clients' performance in many applications is stochastic and hard to estimate. Properly handling this stochasticity motivates this research, as it is not well addressed in pioneering literature. In this paper, we focus on cross-device FL and propose a multi-level FL architecture under the real scenarios. Considering the two properties of clients' situations: uncertainty, correlation, we propose FL Incentive Mechanism based on Portfolio theory (FL-IMP). As far as we are aware, this is the pioneering application of portfolio theory to incentive mechanism design aimed at resolving FL resource allocation problem. In order to more accurately reflect practical FL scenarios, we introduce the Federated Learning Agent-Based Model (FL-ABM) as a means of simulating autonomous clients. FL-ABM enables us to gain a deeper understanding of the factors that influence the system's outcomes. Experimental evaluations of our approach have extensively validated its effectiveness and superior performance in comparison to the benchmark methods.

cs.GT↗

arXiv2021

High performance distributed feedback quantum dot lasers with laterally coupled dielectric grating

The combination of grating-based frequency-selective optical feedback mechanisms, such as distributed feedback (DFB) or distributed Bragg reflector (DBR) structures, with quantum dot (QD) gain materials is a main approach towards ultra-high-performance semiconductor lasers for many key novel applications, either as stand-alone sources or as on-chip sources in photonic integrated circuits. However, the fabrication of conventional buried Bragg grating structures on GaAs, GaAs/Si, GaSb and other material platforms have been met with major material regrowth difficulties. We report a novel and universal approach of introducing laterally coupled dielectric Bragg gratings to semiconductor lasers that allows highly controllable, reliable and strong coupling between the grating and the optical mode. We implement such a grating structure in a low-loss amorphous silicon material alongside GaAs lasers with InAs/GaAs QD gain layers. The resulting DFB laser arrays emit at pre-designed 0.8 THz LWDM frequency intervals in the 1300 nm band with record performance parameters, including side mode suppression ratios as high as 52.7 dB, continuous-wave output power of 27.7 mW (room-temperature) and 10 mW (at 70°C), and ultra-low relative intensity noise (RIN) of < -165 dB/Hz (2.5-25 GHz). The devices are also capable of operating isolator-free under very high external reflection levels of up to -12.3 dB whilst maintaining the high spectral and ultra-low RIN qualities. These results validate the novel laterally coupled dielectric grating as a technologically superior and potentially cost-effective approach for fabricating DFB and DBR lasers free of their semiconductor material constraints, thus universally applicable across different material platforms and wavelength bands.

physics.optics↗

arXiv2020

Lossless Image Compression through Super-Resolution

We introduce a simple and efficient lossless image compression algorithm. We store a low resolution version of an image as raw pixels, followed by several iterations of lossless super-resolution. For lossless super-resolution, we predict the probability of a high-resolution image, conditioned on the low-resolution input, and use entropy coding to compress this super-resolution operator. Super-Resolution based Compression (SReC) is able to achieve state-of-the-art compression rates with practical runtimes on large datasets. Code is available online at https://github.com/caoscott/SReC.

eess.IV↗