Search for: All records

Abstract Protein language models (pLMs) have been widely adopted for various protein and peptide-related downstream tasks and demonstrated promising performance. However, short peptides are significantly underrepresented in commonly used pLM training datasets. For example, only 2.8% of sequences in the UniProt Reference Cluster (UniRef) contain fewer than 50 residues, which potentially limits the effectiveness of pLMs for peptide-specific applications. Here, we present PepBERT, a lightweight and efficient peptide language model specifically designed for encoding peptide sequences. Two versions of the model—PepBERT-large (4.9 million parameters) and PepBERT-small (1.86 million parameters)—were pretrained from scratch using four custom peptide datasets and evaluated on nine peptide-related downstream prediction tasks. Both PepBERT models achieved performance superior to or comparable to the benchmark model, ESM-2 with 7.5 million parameters, on 8 out of 9 datasets. Overall, PepBERT provides a compact yet effective solution for generating high-quality peptide representations for downstream applications. By enabling more accurate representation and prediction of bioactive peptides, PepBERT can accelerate the discovery of food-derived bioactive peptides with health-promoting properties, supporting the development of sustainable functional foods and value-added utilization of food processing by-products. The datasets, source codes, pretrained models, and tutorials for the usage of PepBERT are available athttps://github.com/dzjxzyd/PepBERT. 

Mixture-of-Experts (MoE) architectures achieve efficient LLM inference through sparse activation, but this transforms static computation graphs into dynamic, token-dependent control flows that create new hardware security vulnerabilities. We present a cross-layer fault injection framework targeting the control-flow primitives of MoE execution. Using Qwen1.5-MoE as a case study, we show that the gating mechanism is a structural single-point-of-failure: a single bit-flip in the router logits collapses reasoning performance to 0%, while equivalent faults in dense models produce only minor numerical noise. We validate two complementary defenses: (1) gate-level Triple Modular Redundancy (TMR) that triplicates routing logic and applies majority voting, recovering discriminative accuracy (BBQ: 41%, toward baseline 43%) with 0.04% parameter overhead; and (2) a Validity Checker fallback that detects anomalous logits and routes tokens through the shared expert only, preventing catastrophic collapse (BBQ: 25% → 47%) at 44% reduced latency. 

RTL generation is more than code synthesis. Designs must be syntactically valid, synthesizable, correct, hardware-efficient. SOTA evaluations stop at functional correctness and do not measure synthesis and implementation quality. This paper evaluates 32 language models on 202 Verilog tasks from VerilogEval and RTLLM using the Hardware Quality Index (HQI) that combines post-synthesis area, delay, and warnings related to expert references in a Nangate45 45 nm flow. Three performance regimes emerge: 14 frontier models achieve HQI > 66, led by Gemini-3-Pro at 87.5% coverage and 85.1 HQI; 15 models cluster 43–66 HQI; 3 are below 43. Gap between best-of-five capability and single-attempt quality spans 3.7–22.1 HQI points, limiting integration into agentic pipelines. A taxonomy of 195 synthesis failures reveals systematic divergence: proprietary models fail late through elaboration errors and synthesis timeout; open models fail early often due to missing module wrappers and non-synthesizable constructs, a pattern consistent with training corpora skewed toward simulation over synthesis-grade RTL. 

Protein language models (pLMs) have been widely adopted for various protein and peptide related prediction tasks and demonstrated promising performance. However, short peptides are substantially underrepresented in commonly used high-capacity pLM training datasets, which may limit model effectiveness for peptide-specific tasks. This study aimed to deliver a lightweight and efficient peptide language model for accurate peptide representation and downstream prediction. To achieve this, we constructed PepBERT, a peptide-specific language model pretrained from scratch on four custom peptide datasets comprising between 1.97 and 19.2 million peptide sequences. Two versions, PepBERT-large (4.9 million parameters) and PepBERT-small (1.86 million parameters), were developed and comprehensively evaluated on nine peptide prediction tasks. Both variants achieved best performance on the UniParc dataset with 19.2 million peptide sequences, and PepBERT outperformed or matched the benchmark model ESM-2 (7.5 million parameters) on eight of nine tasks, while requiring substantially fewer computational resources. In terms of inference efficiency, PepBERT-large and PepBERT-small achieved 16 × and 25 × faster inference speeds, respectively, compared withESM-2. These results demonstrate that PepBERT effectively captures the sequential dependencies and positional context within peptide sequences and generates expressive and generalizable peptide embeddings across diverse bioactivity prediction tasks with significantly lower computational cost. By improving peptide representation and downstream prediction, PepBERT can accelerate the discovery of food-derived bioactive peptides and support sustainable functional food innovation. All datasets, pretrained models, and codes are publicly available at https://github.com/dzjxzyd/PepBERT. 

We propose a malware-centric blacklisting framework for side-channel security monitoring that identifies the type of malware present on embedded and IoT devices using electromagnetic (EM) signatures. In contrast to traditional reference model-based monitoring that primarily learns benign behavior and identifies deviations, our approach employs an Inverse One-Class Classification strategy grounded in the Complexity Hypothesis. By utilizing a 1D CNN-Autoencoder with a constrained bottleneck, we model the reconstruction residual as a Signal Complexity Indicator. This allows the framework to distinguish between high-entropy attack signals and low-dimensional benign executions without exhaustive benign profiling. At run time, the monitor analyzes short side-channel windows and outputs either a matched malware type or a principled no-match decision. We evaluate the proposed framework on the Arty-A7 100t FPGA platform using the NIST SARD Juliet Test Suite. Experimental results demonstrate that SC2V achieves an attack detection accuracy of 94.34% and robust malware-type identification, providing actionable, type-aware alerts for incident response. 

Embedded systems are increasingly targeted by sophisticated malware, but resource constraints and heterogeneous hardware make conventional defenses ineffective against stealthy attacks. We present RuntimeShield , a runtime anomaly detection framework that leverages the on-chip power distribution network as a hardware trust anchor for embedded system security. Lightweight time-to-digital converter sensors embedded in the power distribution network capture high-resolution power traces. These measurements are analyzed by EchoDropNet, an autocorrelation-enhanced, multi-scale CNN designed to extract stable features from weak and noisy side-channel signatures. We implement RuntimeShield on RISC-V, ARM (A9/A53/R5), and MicroBlaze and evaluate it on 25 BEEBS benchmarks plus open-source workloads. various attacks, RuntimeShield achieves 96.85% runtime detection accuracy with low overhead, demonstrating a practical path to on-device protection for embedded systems. 

We present BlindMarket, an end-to-end zero-trust distribution framework for hardware IP cores that operates without relying on any trusted third party. BlindMarket allows two parties, the IP user and the IP vendor, to complete an IP trading process with strong guarantees of verifiability and confidentiality before the transaction, and then traceability after. We propose verification heuristics and adapt the cone of influencebased design pruning to overcome the limited scalability common to cryptographic protocols and the hardness of the underlying hardware verification. We systematically evaluate our framework on a diverse set of real-world hardware benchmarks. The results demonstrate that BlindMarket successfully verifies 12 out of 13 IP designs and achieves substantial performance improvements enabled by design pruning and control-flow guided heuristics. 

Large Language Models (LLMs) demonstrate impressive capabilities across many applications but remain vulnerable to jailbreak attacks, which elicit harmful or unintended content. While model fine-tuning is an option for safety alignment, it is costly and prone to catastrophic forgetting. Prompt optimization has emerged as a promising alternative, yet existing prompt-based defenses typically rely on static modifications (e.g., fixed prefixes or suffixes) that cannot adapt to diverse and evolving attacks.We propose Dynamic Deep Prompt Optimization (DDPO), the first jailbreak defense based on deep prompt optimization. DDPO uses the target LLM's own intermediate layers as feature extractors to dynamically generate defensive embeddings via a lightweight multilayer perceptron. These tailored embeddings are then injected into a subsequent intermediate layer, enabling an input-dependent defense without modifying the LLM's weights. This design ensures high adaptability with minimal computational overhead.Experiments on a diverse set of models and attacks demonstrate that DDPO significantly outperforms static prompt optimization methods, particularly on weakly aligned models and when handling semantically ambiguous benign prompts, successfully distinguishing them from genuinely harmful requests.