Search

Diffusion models have achieved great success in a range of tasks, such as image synthesis and molecule design. As such successes hinge on large-scale training data collected from diverse sources, the trustworthiness of these collected data is hard to control or audit. In this work, we aim to explore the vulnerabilities of diffusion models under potential training data manipulations and try to answer: How hard is it to perform Trojan attacks on well-trained diffusion models? What are the adversarial targets that such Trojan attacks can achieve? To answer these questions, we propose an effective Trojan attack against diffusion models, TrojDiff, which optimizes the Trojan diffusion and generative processes during training. In particular, we design novel transitions during the Trojan diffusion process to diffuse adversarial targets into a biased Gaussian distribution and propose a new parameterization of the Trojan generative process that leads to an effective training objective for the attack. In addition, we consider three types of adversarial targets: the Trojaned diffusion models will always output instances belonging to a certain class from the in-domain distribution (In-D2D attack), out-of-domain distribution (Out-D2D-attack), and one specific instance (D2I attack). We evaluate TrojDiff on CIFAR-10 and CelebA datasets against both DDPM and DDIM diffusion models. We show that TrojDiff always achieves high attack performance under different adversarial targets using different types of triggers, while the performance in benign environments is preserved. The code is available at https://github.com/chenweixin107/TrojDiff., Comment: CVPR2023

Generative Pre-trained Transformer (GPT) models have exhibited exciting progress in capabilities, capturing the interest of practitioners and the public alike. Yet, while the literature on the trustworthiness of GPT models remains limited, practitioners have proposed employing capable GPT models for sensitive applications to healthcare and finance - where mistakes can be costly. To this end, this work proposes a comprehensive trustworthiness evaluation for large language models with a focus on GPT-4 and GPT-3.5, considering diverse perspectives - including toxicity, stereotype bias, adversarial robustness, out-of-distribution robustness, robustness on adversarial demonstrations, privacy, machine ethics, and fairness. Based on our evaluations, we discover previously unpublished vulnerabilities to trustworthiness threats. For instance, we find that GPT models can be easily misled to generate toxic and biased outputs and leak private information in both training data and conversation history. We also find that although GPT-4 is usually more trustworthy than GPT-3.5 on standard benchmarks, GPT-4 is more vulnerable given jailbreaking system or user prompts, potentially due to the reason that GPT-4 follows the (misleading) instructions more precisely. Our work illustrates a comprehensive trustworthiness evaluation of GPT models and sheds light on the trustworthiness gaps. Our benchmark is publicly available at https://decodingtrust.github.io/.

Two pyrene-degrading strains, Pseudomonas aeruginosa PA06 and Achromobacter sp. AC15 were co-incubated in equal proportions as a microbiological consortium and could enhance the degradation of pyrene. The enzymatic activities of the catechol 1,2-dioxygenase (C12O) and 2,3-dioxygenase activities (C23O) were produced complementary expression by P. aeruginosa PA06 and Achromobacter sp. AC15, respectively. Meanwhile, results showed that pyrene degradation was sufficiently promoted in the presence of sodium citrate as a co-metabolic carbon source, likely a result of enhanced biomass and biosurfactant production. The optimized dosage and ideal initial pHs were 1.4 g L−1 and 5.5, respectively. We also analyzed the rate constant of pyrene degradation, cell growth, and enzyme activity. Results show that P. aeruginosa PA06 had a better effect than Achromobacter sp. AC15 in bacterial growth. However, the C23O or C12O activity produced by Achromobacter sp. AC15 continued at a similar or even faster than that of P. aeruginosa PA06. The mixed bacteria had a better effect than any single bacteria, suggesting the strains worked synergistically to enhance the degradation efficiency. In the co-metabolism system of 600 mg/L pyrene and 1.4 g/L sodium citrate, pyrene degradation reached 74.6%, was 1.57 times, 2.06 times, and 3.89 times that of the mix-culture strains, single PA06 and single AC15 without sodium citrate, respectively. Overall, these findings are valuable as a potential tool for the bioremediation of high-molecular-weight PAHs.

Additional file 1: Table S1. HAV reference variantsand GenBank ascension number

A face tracking sign-in system based on LBP algorithm was designed. The system is integrated by OPENMV module, steering engine, HMI and wireless communication module, etc. These modules cooperate with each other to construct a face tracking sign-in system based on LBP feature algorithm. Haar-cascade detection algorithm is adopted to detect face images; LBP feature algorithm is used to extract face features; SVM support vector machine is used to classify the extracted feature vectors. The HMI serial screen is adopted for human-computer interaction, and the sign-in result is fed back to the mobile phone app through the wireless communication module. The main advantage of the system is that the system can intelligently find and track the position of the face through the two-axis gimbal combined with the image data. Improve the efficiency of face recognition and optimize the characteristics of human-computer interaction. The system is easy to implement, and has the advantages of high recognition rate and efficiency.

Bacillus pseudomycoides BS6, a bacterial strain isolated from edible oil contaminated soil, was identified to be a potent producer of biosurfactant using soybean oil waste as the sole source of carbon and energy and showed good interfacial characteristics. Crude biosurfactants were obtained from cell-free culture supernatant and was further purified by column chromatography on reverse-phase silica gel, ultra-filtration, acidification and lyophilisation. Thin-layer chromatography (TLC) and Fourier Transform Infrared Spectroscopy (FT-IR) analyses showed that the biosurfactant produced by strain BS6 was lipopeptide. The chemical structure of the purified lipopeptide was analyzed by GC-MS, amino acid analyzer and LTQ Orbitrap Elite combination mass spectrometer. It was inferred that the compound was a novel cyclic lipopeptide with a long-chain fatty acid 3-OH-C 18 and a peptide comprised of five amino acids. The molar concentration of glutamic acid, methionine and lysine in the peptide chain was 19.36%, 56.32%, and 19.61%, respectively. The critical micelle concentration of the purified lipopeptide was only 56 mg l −1 and could reduce the surface tension of water from 71.6 mN m −1 to 30.2 mN m −1 . Its emulsification index against five plant oils and two hydrocarbons was approximately 62.8–94.2%. All these characteristics support bioconversion by BS6 is a promising and commercial way of recycling waste cooking oil and bioremediation in petroleum-contaminated sites.

An Achromobacter species strain AC15 was isolated from mangrove soil and identified as a potent producer of biosurfactant. It could utilize high concentrations of pyrene as the sole source of carbon and energy. Results showed that AC15 degraded about 40% of the pyrene, which was provided at an initial concentration of 300 mg L−1, and reduced surface tension of the culture medium from 67.2 to 33.2 mN m−1 after 14 d of incubation using its good emulsification property on the culture medium. The main chemical component of partially purified biosurfactant, namely BS15, was analyzed by gas chromatography-mass spectrometry, amino acid analysis, and LTQ Orbitrap Elite mass spectrometry. Results showed that BS15 was a novel linear lipopeptide with a long-chain C16 fatty acid and peptide comprised of four amino acid residues. The molecular structure was determined as CH3-(CH2)14–CO–O-Leu-Asp-Lys-Ser. The solubility of pyrene in water improved 1.5–1.9 times in the presence of 40–60 mg L−1 of BS15. This work demonstrated that strain AC15 and its biosurfactant BS15 could be potential candidates in further applications of bioremediation of high-molecular weight polycyclic aromatic hydrocarbons.

In order to obtain radicals of a Г-ring determined by strong nilpotency, nilpotency, and strongly nil property, we introduce a general method. Let P be an arbitrary property of Γ-rings. For any Γ-ring we construct a quasi-P radical determined by P. If P is homomorphically closed, then this quasi-P radical can be constructed by transfinite induction. We next study the relations of radicals among the quasi-strongly nil radical and other radicals of Г-rings. In view of these relations, we see that the quasi-strongly nil radical is a new radical in the radical theory of Г-rings.

Let P be an arbitrary property of rings and homomorphically closed. For an arbitrary ring A we construct quasi-P radical QP(A) with transfinite induction, and give another characterization of quasi-P radical: QP(A) = ∩ Ia ∣Ia. is an ideal of A, and A/I a contains no non-zero P-ideal.

A hierarchical lamellar porous carbon material was prepared with fish scale using a natural template. Electric double layer capacitors electrodes prepared from this kind of porous carbon exhibited exceptional ration ability which demonstrated that fish scale is a promising candidate precursor to prepare low cost but high performance porous carbon material.

Pygmaeocine E, a diterpenoid with a new carbon skeleton, was isolated from roots of Pygmaeopremna herbacea . Its structure was established by spectroscopic methods and verified by X-ray crystallography.