High-performance neural language models have obtained state-of-the-art results on a wide range of Natural Language Processing (NLP) tasks. However, results for common benchmark datasets often do not reflect model reliability and robustness when applied to noisy, real-world data. In this study, we design and implement various types of character-level and word-level perturbation methods to simulate realistic scenarios in which input texts may be slightly noisy or different from the data distribution on which NLP systems were trained. Conducting comprehensive experiments on different NLP tasks, we investigate the ability of high-performance language models such as BERT, XLNet, RoBERTa, and ELMo in handling different types of input perturbations. The results suggest that language models are sensitive to input perturbations and their performance can decrease even when small changes are introduced. We highlight that models need to be further improved and that current benchmarks are not reflecting model robustness well. We argue that evaluations on perturbed inputs should routinely complement widely-used benchmarks in order to yield a more realistic understanding of NLP systems robustness.

We study the Bayesian inverse problem of learning a linear operator on a Hilbert space from its noisy pointwise evaluations on random input data. Our framework assumes that this target operator is self-adjoint and diagonal in a basis shared with the Gaussian prior and noise covariance operators arising from the imposed statistical model and is able to handle target operators that are compact, bounded, or even unbounded. We establish posterior contraction rates with respect to a family of Bochner norms as the number of data tend to infinity and derive related lower bounds on the estimation error. In the large data limit, we also provide asymptotic convergence rates of suitably defined excess risk and generalization gap functionals associated with the posterior mean point estimator. In doing so, we connect the posterior consistency results to nonparametric learning theory. Furthermore, these convergence rates highlight and quantify the difficulty of learning unbounded linear operators in comparison with the learning of bounded or compact ones. Numerical experiments confirm the theory and demonstrate that similar conclusions may be expected in more general problem settings.

The artificial intelligence systems used by image recognition tools, such as those that certain connected cars use to identify street signs, can be tricked to make an incorrect identification by a low-cost but effective attack using a camera, a projector and a PC, according to Purdue University researchers. A research paper describes an Optical Adversarial Attack, or OPAD, which uses a projector to project calculated patterns that alter the appearance of the 3D objects to AI-based image recognition systems. The paper will be presented in October at an ICCV 2021 Workshop. In an experiment, a pattern was projected onto a stop sign, causing the image recognition to read the sign as a speed limit sign instead. The researchers say this attack method could also work with image recognition tools in applications ranging from military drones to facial recognition systems, potentially undermining their reliability.

RNA is distinct among large biomolecules in that it has both informational coding ability, carried in its sequence, and the ability to form complex three-dimensional structures that can have catalytic and regulatory roles. The information-carrying component is widely appreciated. The pattern of base pairing—the first level of RNA structure—can be experimentally assessed and modeled with impressive accuracy ([ 1 ][1], [ 2 ][2]). By contrast, our understanding of the extent and roles of complex three-dimensional RNA structures remains rudimentary. RNA viral genomes are rich in motifs with complex three-dimensional structures with regulatory functions ([ 3 ][3]), and evidence increasingly supports the hypothesis that functional RNA structures are ubiquitous in organisms ranging from bacteria to humans. However, developing and testing hypotheses about the roles of RNA structure have been hindered by the inability to identify and model these structures. On page 1047 of this issue, Townshend et al. ([ 4 ][4]) report a machine-learning strategy for identifying native-like RNA folds. Nearly all RNAs that form well-understood complex structures fall into a small number of classes: the ribosomal RNAs, the large and small ribozymes that catalyze RNA cleavage, bacterial riboswitches, and regulatory elements encoded by RNA viruses. Thus, there are limited examples for guiding identification and modeling of RNAs with complex three-dimensional structures. There are only four major RNA nucleotides, and the interactions that govern base pairing and simple helix formation are well understood. Once formed, RNA helices (secondary structure) often assemble as fairly rigid elements that interact hierarchically to form more complicated structures (tertiary structure) (see the figure). Despite these simplifying features, the modeling of complex RNA structures has proven to be difficult. The RNA-Puzzles community exercise ([ 5 ][5], [ 6 ][6]) has been instrumental in illuminating the challenges involved: Groups try to predict an RNA structure from its sequence before learning the solved structure. Several rounds of RNA-Puzzles have revealed important themes. No single method consistently yields the best models, although certain approaches have better records than others, and most approaches are getting better. The best agreement tends to result when experimental or homology-based information is incorporated into the computational modeling. However, the median accuracy for small RNAs, with complex tertiary folds but without a close known homolog, has stayed stubbornly stuck in a range of ∼15- to 20-Å root mean square deviation [(RMSD) a measure of the similarity between known and modeled structures]. This agreement is much poorer than that now achieved for protein structures by machine learning ([ 7 ][7]), where native-like folds (∼2-Å RMSD or less) are achieved. Modeled RNA structures thus often recapitulate the overall fold of a target RNA but do not consistently reveal details of the tertiary structure. Current methods are not likely to be useful for applications such as understanding the biological mechanism of a structure or for designing ligands (or drugs) that modulate RNA function. ![Figure][8] RNA structure RNA molecules have multiple levels of structure and ability to encode information. The sequence of RNA is readily determined. RNA secondary structure can now be elucidated with high levels of accuracy using approaches that meld computational energy minimization with experimental per-nucleotide chemical probing information. Townshend et al. developed a deep neural network that can identify models that best represent the native tertiary state, taking a step toward modeling three-dimensional RNA structure. GRAPHIC: C. BICKEL/ SCIENCE The Atomic Rotationally Equivalent Scorer (ARES) approach of Townshend et al. is a deep neural network, a form of machine learning, and did not initially include preconceived notions of RNA structure. Indeed, the ARES framework is not specific to RNA and can be applied to other problems in molecular structure. Instead, ARES was given a small set of motifs with known RNA structure plus a large number of alternative (incorrect) variations of these same structures. ARES parameters were adjusted so that the program learned the functional and geometric arrangements of each atom and how these elements are positioned relative to each other. Layers in the neural network compute features from finer to coarser scales to recognize base pairs, helices, and more-complex structures. For example, ARES learned patterns of base pairing, the optimal geometry for RNA helices, and a subset of noncanonical tertiary motifs without being provided explicit information about these features of RNA structure. Although ARES was trained on very simple RNA systems, the resulting ARES scoring function was able to predict structures of more complex RNAs, on average, to roughly a 12-Å RMSD. This degree of accuracy represents an overall improvement of ∼4 Å over prior scoring methods. ARES is still short of the level consistent with atomic resolution or sufficient to guide identification of key functional sites or drug discovery efforts, but Townshend et al. have achieved notable progress in a field that has proven recalcitrant to transformative advances. There are three fundamental challenges for modeling complex RNA three-dimensional structures: generating reasonable structures that may represent a biological state, accurately scoring or identifying models that best represent the correct native state, and using these hopefully accurate models to discover new functional motifs and to develop hypotheses regarding the mechanisms by which RNAs with complex three-dimensional structures regulate biological processes. The ARES machine-learning approach addressed the second of these three challenges: Candidate structures still need to be generated for evaluation by ARES. With further development, deep learning strategies hold promise for creating new scoring functions that can guide structure generation in ways that might yield near-native structures. Another important goal is to use a machine-learning strategy to identify regions in large RNAs most likely to fold into three-dimensional structures. Current computational-only algorithms are not able to predict the pattern of base pairing in large RNAs accurately, even though base pairs are simpler to predict than tertiary structure. However, secondary structures for large RNAs are routinely modeled to high accuracies by incorporating experimental information. New, efficiently executed experiments are now being developed that measure features of RNA tertiary structures. Another frontier, analogous to recent advances in secondary structure modeling, would thus be to incorporate experimental information into machine-learning strategies for modeling RNA tertiary structure. Large-scale investigation of RNA structure to date, primarily focused on RNA secondary structure, has revealed several core principles. One is that the existence of regions within large RNAs with complex, higher-order structure is unremarkable. When these base pairing and tertiary structures affect biological functions, they create “an RNA structure code” with pervasive effects on gene regulatory circuits. Additionally, every RNA likely has a distinct structural personality, which implies that there are numerous ways by which RNA structure tunes the underlying function of an RNA. At the level of secondary structure, such tuning RNA structures tend to function like switches and attenuators that modulate binding by RNA and protein ligands ([ 8 ][9]–[ 11 ][10]). Finally, characterization of well-determined RNA secondary structures often leads to identification of centers of new biology. As it becomes possible to measure, (deeply) learn, and predict the details of the tertiary RNA structure-ome, diverse new discoveries in biological mechanisms await. 1. [↵][11]1. E. J. Strobel et al ., Nat. Rev. Genet. 19, 615 (2018). [OpenUrl][12][CrossRef][13][PubMed][14] 2. [↵][15]1. K. M. Weeks , Acc. Chem. Res. 54, 2502 (2021). [OpenUrl][16][CrossRef][17] 3. [↵][18]1. Z. A. Jaafar, 2. J. S. Kieft , Nat. Rev. Microbiol. 17, 110 (2019). [OpenUrl][19][CrossRef][20] 4. [↵][21]1. R. J. L. Townshend et al ., Science 373, 1047 (2021). [OpenUrl][22][Abstract/FREE Full Text][23] 5. [↵][24]1. J. A. Cruz et al ., RNA 18, 610 (2012). [OpenUrl][25][Abstract/FREE Full Text][26] 6. [↵][27]1. Z. Miao et al ., RNA 26, 982 (2020). [OpenUrl][28][Abstract/FREE Full Text][29] 7. [↵][30]1. E. Pennisi , Science 373, 262 (2021). [OpenUrl][31][Abstract/FREE Full Text][32] 8. [↵][33]1. D. Long et al ., Nat. Struct. Mol. Biol. 14, 287 (2007). [OpenUrl][34][CrossRef][35][PubMed][36][Web of Science][37] 9. 1. M. Kertesz et al ., Nat. Genet. 39, 1278 (2007). [OpenUrl][38][CrossRef][39][PubMed][40][Web of Science][41] 10. 1. D. Dominguez et al ., Mol. Cell 70, 854 (2018). [OpenUrl][42][CrossRef][43][PubMed][44] 11. [↵][45]1. A. M. Mustoe et al ., Biochemistry 57, 3537 (2018). [OpenUrl][46][CrossRef][47] Acknowledgments: The author’s laboratory is supported by the US National Institutes of Health and National Science Foundation. The author is an advisor to and holds equity in Ribometrix. [1]: #ref-1 [2]: #ref-2 [3]: #ref-3 [4]: #ref-4 [5]: #ref-5 [6]: #ref-6 [7]: #ref-7 [8]: pending:yes [9]: #ref-8 [10]: #ref-11 [11]: #xref-ref-1-1 "View reference 1 in text" [12]: {openurl}?query=rft.jtitle%253DNat.%2BRev.%2BGenet.%26rft.volume%253D19%26rft.spage%253D615%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fs41576-018-0034-x%26rft_id%253Dinfo%253Apmid%252F30054568%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [13]: /lookup/external-ref?access_num=10.1038/s41576-018-0034-x&link_type=DOI [14]: /lookup/external-ref?access_num=30054568&link_type=MED&atom=%2Fsci%2F373%2F6558%2F964.atom [15]: #xref-ref-2-1 "View reference 2 in text" [16]: {openurl}?query=rft.jtitle%253DAcc.%2BChem.%2BRes.%26rft.volume%253D54%26rft.spage%253D2502%26rft_id%253Dinfo%253Adoi%252F10.1021%252Facs.accounts.1c00118%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [17]: /lookup/external-ref?access_num=10.1021/acs.accounts.1c00118&link_type=DOI [18]: #xref-ref-3-1 "View reference 3 in text" [19]: {openurl}?query=rft.jtitle%253DNat.%2BRev.%2BMicrobiol.%26rft.volume%253D17%26rft.spage%253D110%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fs41579-018-0117-x%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [20]: /lookup/external-ref?access_num=10.1038/s41579-018-0117-x&link_type=DOI [21]: #xref-ref-4-1 "View reference 4 in text" [22]: {openurl}?query=rft.jtitle%253DScience%26rft.stitle%253DScience%26rft.aulast%253DTownshend%26rft.auinit1%253DR.%2BJ.%2BL.%26rft.volume%253D373%26rft.issue%253D6558%26rft.spage%253D1047%26rft.epage%253D1051%26rft.atitle%253DGeometric%2Bdeep%2Blearning%2Bof%2BRNA%2Bstructure%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fscience.abe5650%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [23]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEzOiIzNzMvNjU1OC8xMDQ3IjtzOjQ6ImF0b20iO3M6MjI6Ii9zY2kvMzczLzY1NTgvOTY0LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ== [24]: #xref-ref-5-1 "View reference 5 in text" [25]: {openurl}?query=rft.jtitle%253DRNA%26rft_id%253Dinfo%253Adoi%252F10.1261%252Frna.031054.111%26rft_id%253Dinfo%253Apmid%252F22361291%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [26]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoicm5hIjtzOjU6InJlc2lkIjtzOjg6IjE4LzQvNjEwIjtzOjQ6ImF0b20iO3M6MjI6Ii9zY2kvMzczLzY1NTgvOTY0LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ== [27]: #xref-ref-6-1 "View reference 6 in text" [28]: {openurl}?query=rft.jtitle%253DRNA%26rft_id%253Dinfo%253Adoi%252F10.1261%252Frna.075341.120%26rft_id%253Dinfo%253Apmid%252F32371455%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [29]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoicm5hIjtzOjU6InJlc2lkIjtzOjg6IjI2LzgvOTgyIjtzOjQ6ImF0b20iO3M6MjI6Ii9zY2kvMzczLzY1NTgvOTY0LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ== [30]: #xref-ref-7-1 "View reference 7 in text" [31]: {openurl}?query=rft.jtitle%253DScience%26rft.stitle%253DScience%26rft.aulast%253DPennisi%26rft.auinit1%253DE.%26rft.volume%253D373%26rft.issue%253D6552%26rft.spage%253D262%26rft.epage%253D263%26rft.atitle%253DProtein%2Bstructure%2Bprediction%2Bnow%2Beasier%252C%2Bfaster%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fscience.373.6552.262%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [32]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEyOiIzNzMvNjU1Mi8yNjIiO3M6NDoiYXRvbSI7czoyMjoiL3NjaS8zNzMvNjU1OC85NjQuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9 [33]: #xref-ref-8-1 "View reference 8 in text" [34]: {openurl}?query=rft.jtitle%253DNature%2Bstructural%2B%2526%2Bmolecular%2Bbiology%26rft.stitle%253DNat%2BStruct%2BMol%2BBiol%26rft.aulast%253DNollmann%26rft.auinit1%253DM.%26rft.volume%253D14%26rft.issue%253D4%26rft.spage%253D287%26rft.epage%253D294%26rft.atitle%253DPotent%2Beffect%2Bof%2Btarget%2Bstructure%2Bon%2BmicroRNA%2Bfunction.%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fnsmb1226%26rft_id%253Dinfo%253Apmid%252F17401373%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [35]: /lookup/external-ref?access_num=10.1038/nsmb1226&link_type=DOI [36]: /lookup/external-ref?access_num=17401373&link_type=MED&atom=%2Fsci%2F373%2F6558%2F964.atom [37]: /lookup/external-ref?access_num=000245469600014&link_type=ISI [38]: {openurl}?query=rft.jtitle%253DNature%2Bgenetics%26rft.stitle%253DNat%2BGenet%26rft.aulast%253DKertesz%26rft.auinit1%253DM.%26rft.volume%253D39%26rft.issue%253D10%26rft.spage%253D1278%26rft.epage%253D1284%26rft.atitle%253DThe%2Brole%2Bof%2Bsite%2Baccessibility%2Bin%2BmicroRNA%2Btarget%2Brecognition.%26rft_id%253Dinfo%253Adoi%252F10.1038%252Fng2135%26rft_id%253Dinfo%253Apmid%252F17893677%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [39]: /lookup/external-ref?access_num=10.1038/ng2135&link_type=DOI [40]: /lookup/external-ref?access_num=17893677&link_type=MED&atom=%2Fsci%2F373%2F6558%2F964.atom [41]: /lookup/external-ref?access_num=000249737400026&link_type=ISI [42]: {openurl}?query=rft.jtitle%253DMol.%2BCell%26rft.volume%253D70%26rft.spage%253D854%26rft_id%253Dinfo%253Adoi%252F10.1016%252Fj.molcel.2018.05.001%26rft_id%253Dinfo%253Apmid%252F29883606%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [43]: /lookup/external-ref?access_num=10.1016/j.molcel.2018.05.001&link_type=DOI [44]: /lookup/external-ref?access_num=29883606&link_type=MED&atom=%2Fsci%2F373%2F6558%2F964.atom [45]: #xref-ref-11-1 "View reference 11 in text" [46]: {openurl}?query=rft.jtitle%253DBiochemistry%26rft.volume%253D57%26rft.spage%253D3537%26rft_id%253Dinfo%253Adoi%252F10.1021%252Facs.biochem.8b00395%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [47]: /lookup/external-ref?access_num=10.1021/acs.biochem.8b00395&link_type=DOI

New research has found that artificial intelligence (AI) analyzing medical scans can identify the race of patients with an astonishing degree of accuracy, while their human counterparts cannot. With the Food and Drug Administration (FDA) approving more algorithms for medical use, the researchers are concerned that AI could end up perpetuating racial biases. They are especially concerned that they could not figure out precisely how the machine-learning models were able to identify race, even from heavily corrupted and low-resolution images. In the study, published on pre-print service Arxiv, an international team of doctors investigated how deep learning models can detect race from medical images. Using private and public chest scans and self-reported data on race and ethnicity, they first assessed how accurate the algorithms were, before investigating the mechanism.

Spear phishing is a social engineering technique targeted towards a targeted individual to divulge confidential information. But creating highly targeted mass spear-phishing emails could take a lot of effort and time. In a recent test conducted by a team of researchers, it was found that they could use Natural Language Processing (NLP) to devise targeted phishing emails. At the end of the research, the team revealed that AI/ML could be used to develop spear-phishing campaigns at a devastating scale. In the recently held Black Hat Defcon security conference in Las Vegas, a team of researchers hailing from the Singapore Government Technology Agency presented the results of their AI/ML generated phishing email test.

The ghost of Edward Teller must have been doing the rounds between members of the National Commission on Artificial Intelligence. The father of the hydrogen bomb was never one too bothered by the ethical niggles that came with inventing murderous technology. It was not, for instance, "the scientist's job to determine whether a hydrogen bomb should be constructed, whether it should be used, or how it should be used." Responsibility, however exercised, rested with the American people and their elected officials. The application of AI in military systems has plagued the ethicist but excited certain leaders and inventors.

In recent years, graph convolutional networks (GCNs) have emerged rapidly due to their excellent performance in graph data processing. However, recent researches show that GCNs are vulnerable to adversarial attacks. An attacker can maliciously modify edges or nodes of the graph to mislead the model’s classification of the target nodes, or even cause a degradation of the model’s overall classification performance. In this paper, we first propose a black-box adversarial attack framework based on derivative-free optimization (DFO) to generate graph adversarial examples without using gradient and apply advanced DFO algorithms conveniently. Second, we implement a direct attack algorithm (DFDA) using the Nevergrad library based on the framework. Additionally, we overcome the problem of large search space by redesigning the perturbation vector using constraint size. Finally, we conducted a series of experiments on different datasets and parameters. The results show that DFDA outperforms Nettack in most cases, and it can achieve an average attack success rate of more than 95% on the Cora dataset when perturbing at most eight edges. This demonstrates that our framework can fully exploit the potential of DFO methods in node classification adversarial attacks.

Today, artificial intelligence is essential across a wide range of industries, including healthcare, retail, manufacturing, and even government. But there are ethical challenges with AI, and as always, we need to stay vigilant about these issues to make sure that artificial intelligence isn't doing more harm than good. Here are some of the biggest ethical challenges of artificial intelligence. We need data to train our artificial intelligence algorithms, and we need to do everything we can to eliminate bias in that data. The ImageNet database, for example, has far more white faces than non-white faces.

Britain will attempt to move away from European data protection regulations as it overhauls its privacy rules after Brexit, the government has announced. The freedom to chart its own course could lead to an end to irritating cookie popups and consent requests online, said the culture secretary, Oliver Dowden, as he called for rules based on "common sense, not box-ticking". But any changes will be constrained by the need to offer a new regime that the EU deems adequate, otherwise data transfers between the UK and EU could be frozen. A new information commissioner will be put in charge of overseeing the transformation. John Edwards, currently the privacy commissioner of New Zealand, has been named as the government's preferred candidate to replace Elizabeth Denham, whose term in office will end on 31 October after a three-month extension.