With the fast spread of machine learning techniques, sharing and adopting public models become very popular.This gives attackers many new opportunities.In this paper, we propose a trojaning attack on neural networks.As are not intuitive for human to understand, features stealthiness.Deploying trojaned can cause various severe consequences including endangering lives (in applications like autonomous driving).We first inverse network generate general trojan trigger, then retrain model with...

Deep neural networks are vulnerable to Trojan attacks. Existing attacks use visible patterns (e.g., a patch or image transformations) as triggers, which human inspection. In this paper, we propose stealthy and efficient attacks, BppAttack. Based on existing biology literature visual systems, quantization dithering the trigger, making imperceptible changes. It is attack without training auxiliary models. Due small changes made images, it hard inject such triggers during training. To alleviate...

Inspired by the great success of using code coverage as guidance in software testing, a lot neural network criteria have been proposed to guide testing models (e.g., model accuracy under adversarial attacks). However, while monotonic relation between and quality has supported many seminal studies engineering, it remains largely unclear whether similar monotonicity exists quality. This paper sets out answer this question. Specifically, correlation DNN criteria, effects guided example...

Planting maize (Zea mays L.) reasonably densely and adding amounts of appropriate nitrogen fertilizer are essential measures to improve the efficiency yield use. In this study, two planting densities 7.5 × 104 plants ha−1 12.0 were established with varieties DengHai 618 (DH618) XianYu 335 (XY335). Simultaneously, 18 levels application established, including a lack (N0) increments 45 kg up 765 (N765) ha−1. The variables studied included effects rate on characteristics dry matter accumulation...

With machine learning models especially Deep Neural Network (DNN) becoming an integral part of the new intelligent software, tools to support their engineering process are in high demand. Existing DNN debugging either post-training which wastes a lot time training buggy model and requires expertises, or limited on collecting logs without analyzing problem not even fixing them. In this paper, we propose AUTOTRAINER, monitoring automatic repairing tool supports detecting auto five commonly...

Large Language Models (LLMs) have emerged as the new recommendation engines, outperforming traditional methods in both capability and scope, particularly code generation applications. Our research reveals a novel provider bias LLMs, namely without explicit input prompts, these models show systematic preferences for services from specific providers their recommendations (e.g., favoring Google Cloud over Microsoft Azure). This holds significant implications market dynamics societal...

Gully initiation has been a subject of much discussion in the field gully erosion. could be examined by S–A relationship (S = aA − b ) between critical slope head (S) and upstream drainage area (A). erosion plays an important role sediment yield land degradation Dry-hot Valley region Southwest China; however, little research focused on development this region. This study investigated morphology topographic threshold permanent gullies Yuanmou Valley. Data for 36 heads were derived from...

Modern software systems are becoming increasingly complex, relying on a lot of third-party library support. Library behaviors hence an integral part behaviors. Analyzing them is as important analyzing the itself. However, libraries highly challenging due to lack source code, implementation in different languages, and complex optimizations. We observe that many Java functions provide excellent documentation, which concisely describes functionalities functions. develop novel technique can...

Code comments provide abundant information that have been leveraged to help perform various software engineering tasks, such as bug detection, specification inference, and code synthesis. However, developers are less motivated write update comments, making it infeasible error-prone leverage facilitate tasks. In this paper, we propose program analysis systematically derive, refine, propagate comments. For example, by propagation via analysis, can be passed on entities not commented bugs...

With Deep Neural Network (DNN) being integrated into a growing number of critical systems with far-reaching impacts on society, there are increasing concerns their ethical performance, such as fairness. Unfortunately, model fairness and accuracy in many cases contradictory goals to optimize during training. To solve this issue, has been works trying improve by formalizing an adversarial game the level. This approach introduces adversary that evaluates besides its prediction main task,...

Formal program specifications are essential for various software engineering tasks, such as verification, synthesis, code debugging and testing. However, manually inferring formal is not only time-consuming but also error-prone. In addition, it requires substantial expertise. Natural language comments contain rich semantics about behaviors of code, making feasible to infer from comments. Inspired by this, we develop a tool, named C2S, automate the specification synthesis task translating...

Stalk lodging in Huang-Huai-Hai summer maize is a serious problem that reduces yields and precludes the use of mechanical grain harvesting equipment. In order to determine effect nitrogen management on resistance stalk, three application rates 150, 250, 350 kg ha−1 (denoted as N150, N250, N350), different periods (sowing, 6-leaf, 12-leaf, silking) were set. Plant morphology, stalk strength, total carbohydrate, content, yield measured treatments. The results showed rate increased was...

Deep Neural Networks are vulnerable to Trojan (or backdoor) attacks. Reverse-engineering methods can reconstruct the trigger and thus identify affected models. Existing reverse-engineering only consider input space constraints, e.g., size in space. Expressly, they assume triggers static patterns fail detect models with feature such as image style transformations. We observe that both input-space feature-space Trojans associated hyperplanes. Based on this observation, we design a novel method...

Cellular senescence (CS) is closely related to tissue ageing including bone ageing. CS and the senescence-associated secretory phenotype (SASP) have emerged as critical pathogenesis elements of senile osteoporosis. This study aims investigate effect lycopene on osteoporosis.The senescence-accelerated mouse prone 6 (SAMP6) strain mice used osteoporosis model. Daily ingestion for 8 weeks preserves mass, density, strength, microarchitecture in SAMP6 mice. Moreover, these alterations are...

This paper first reviews non-traditional heat exchanger geometry, laser welding, practical issues with microchannel exchangers, and high effectiveness exchangers. Existing exchangers have low material costs, but manufacturing costs. presents a new expanded design accompanying continuous technique for potential low-cost production. Polymer the effectiveness. The discusses one possible joining method - type of welding named "forward conduction welding," used to fabricate prototype. has...

Mature (60-65 days old) male Sprague-Dawley rats received a single i.p. injection of ethane dimethane sulfonate (EDS, 100 mg/kg BW) and were killed at different times from Days 2 to 60 posttreatment. Bands cells enriched in precursor Leydig (PLCs) (LCs) isolated the testis EDS-treated age-matched controls using collagenase digestion-Percoll gradient method. Total RNA extracted PLC LC fractions was subjected reverse transcriptase polymerase chain reaction (RT-PCR) detect estrogen receptor...