Nine language-vision AI models trained on web scrapes with the Contrastive Language-Image Pretraining (CLIP) objective are evaluated for evidence of a bias studied by psychologists: the sexual objectification of girls and women, which occurs when a person's human characteristics are disregarded and the person is treated as a body or a collection of body parts. A first experiment uses standardized images of women from the Sexual OBjectification and EMotion Database, and finds that, commensurate with prior research in psychology, human characteristics are disassociated from images of objectified women: the model's recognition of emotional state is mediated by whether the subject is fully or partially clothed. Embedding association tests (EATs) return significant effect sizes for both anger (d >.8) and sadness (d >.5). A second experiment measures the effect in a representative application: an automatic image captioner (Antarctic Captions) includes words denoting emotion less than 50% as often for images of partially clothed women than for images of fully clothed women. A third experiment finds that images of female professionals (scientists, doctors, executives) are likely to be associated with sexual descriptions relative to images of male professionals. A fourth experiment shows that a prompt of "a [age] year old girl" generates sexualized images (as determined by an NSFW classifier) up to 73% of the time for VQGAN-CLIP (age 17), and up to 40% of the time for Stable Diffusion (ages 14 and 18); the corresponding rate for boys never surpasses 9%. The evidence indicates that language-vision AI models trained on automatically collected web scrapes learn biases of sexual objectification, which propagate to downstream applications.

Machine learning models are now able to convert user-written text descriptions into naturalistic images. These models are available to anyone online and are being used to generate millions of images a day. We investigate these models and find that they amplify dangerous and complex stereotypes. Moreover, we find that the amplified stereotypes are difficult to predict and not easily mitigated by users or model owners. The extent to which these image-generation models perpetuate and amplify stereotypes and their mass deployment is cause for serious concern.

评估了三种最先进的语言和图像AI模型，即剪辑，滑移和BLIP，以证明以前在社会和实验心理学中观察到的偏见：将美国身份等同于白人。使用芝加哥面部数据库（CFD）的自我识别的亚洲，黑人，拉丁裔和白人的标准化图像的嵌入关联测试（eats）表明，白人与集体内词相比，比亚洲，黑色更相关，或拉丁裔/o个人。在评估社会心理学家报道的美国身份的三个核心方面时，单类饮食表明，白人个体的图像与爱国主义和出生在美国更相关，但与心理学的先前发现一致，白人个人是相关的不太可能平等对待所有种族和背景的人。三个下游机器学习任务表明了与白人相关联的偏见。在使用BLIP的视觉问题回答任务中，有97％的白人被确定为美国人，而仅3％的亚洲人。当被问及个人所描绘的生活状态时，该模型在亚洲人中有53％的时间回应中国，但始终具有美国对白人个人的国家。在图像字幕的任务中，Blip评论了亚洲人的种族多达36％的时间，但从未对白人人士进行比赛。最后，使用基于文本的剪辑指导的综合图像发生器（VQGAN）提供了CFD和文本“ American Person”的初始化图像，从而减轻了所有种族个体的肤色（黑人个人的肤色为35％ ，基于像素亮度）。结果表明，语言和图像AI将其等同于美国身份与白人的偏见，并传播到此类模型的下游应用。

语言语料库中的统计规律将众所周知的社会偏见编码为单词嵌入。在这里，我们专注于性别，以全面分析在互联网语料库中训练的广泛使用的静态英语单词嵌入式（Glove 2014，FastText 2017）。使用单类单词嵌入关联测试，我们证明了性别偏见的广泛流行，这些偏见也显示出：（1）与男性与女性相关的单词频率； （b）与性别相关的单词中的言论部分； （c）与性别相关的单词中的语义类别； （d）性别相关的单词中的价，唤醒和优势。首先，就单词频率而言：我们发现，在词汇量中，有1000个最常见的单词与男性相比，有77％的人与男性相关，这是在英语世界的日常语言中直接证明男性默认的证据。其次，转向言论的部分：顶级男性相关的单词通常是动词（例如，战斗，压倒性），而顶级女性相关的单词通常是形容词和副词（例如，奉献，情感上）。嵌入中的性别偏见也渗透到言论部分。第三，对于语义类别：自下而上，对与每个性别相关的前1000个单词的群集分析。与男性相关的顶级概念包括大技术，工程，宗教，体育和暴力的角色和领域；相比之下，顶级女性相关的概念较少关注角色，包括女性特定的诽谤和性内容以及外观和厨房用语。第四，使用〜20,000个单词词典的人类评级，唤醒和主导地位，我们发现与男性相关的单词在唤醒和优势上较高，而与女性相关的单词在价上更高。

Word Embeddings从单词共同发生统计信息中捕获的语言规律学习隐式偏差。通过延长定量单词嵌入中的人类偏差的方法，我们介绍了valnorm，一种新的内在评估任务和方法，以量化人类级字体群体的价值维度与社会心理学。从七种语言（中文，英语，德语，波兰语，葡萄牙语，西班牙语和土耳其语）以及跨越200年的历史英语文本，将Valnorm应用于静态词嵌入式Valnorm在量化非歧视性的非社交组字集的价值方面达到了始终如一的高精度。具体而言，Valnorm实现了r = 0.88的Pearson相关性，用于399个单词的人类判断得分，以建立英语的愉快规范。相比之下，我们使用相同的单词嵌入品测量性别刻板印象，并发现社会偏见因语言而异。我们的结果表明，非歧视性，非社会群组词的价协会代表着七种语言和200多年的广泛共享的协会。

Artificial intelligence and machine learning are in a period of astounding growth. However, there are concerns that these technologies may be used, either with or without intention, to perpetuate the prejudice and unfairness that unfortunately characterizes many human institutions. Here we show for the first time that human-like semantic biases result from the application of standard machine learning to ordinary language-the same sort of language humans are exposed to every day. We replicate a spectrum of standard human biases as exposed by the Implicit Association Test and other well-known psychological studies. We replicate these using a widely used, purely statistical machine-learning model-namely, the GloVe word embedding-trained on a corpus of text from the Web. Our results indicate that language itself contains recoverable and accurate imprints of our historic biases, whether these are morally neutral as towards insects or flowers, problematic as towards race or gender, or even simply veridical, reflecting the status quo for the distribution of gender with respect to careers or first names. These regularities are captured by machine learning along with the rest of semantics. In addition to our empirical findings concerning language, we also contribute new methods for evaluating bias in text, the Word Embedding Association Test (WEAT) and the Word Embedding Factual Association Test (WEFAT). Our results have implications not only for AI and machine learning, but also for the fields of psychology, sociology, and human ethics, since they raise the possibility that mere exposure to everyday language can account for the biases we replicate here.

This paper highlights vulnerabilities of deep learning-driven semantic communications to backdoor (Trojan) attacks. Semantic communications aims to convey a desired meaning while transferring information from a transmitter to its receiver. An encoder-decoder pair that is represented by two deep neural networks (DNNs) as part of an autoencoder is trained to reconstruct signals such as images at the receiver by transmitting latent features of small size over a limited number of channel uses. In the meantime, another DNN of a semantic task classifier at the receiver is jointly trained with the autoencoder to check the meaning conveyed to the receiver. The complex decision space of the DNNs makes semantic communications susceptible to adversarial manipulations. In a backdoor (Trojan) attack, the adversary adds triggers to a small portion of training samples and changes the label to a target label. When the transfer of images is considered, the triggers can be added to the images or equivalently to the corresponding transmitted or received signals. In test time, the adversary activates these triggers by providing poisoned samples as input to the encoder (or decoder) of semantic communications. The backdoor attack can effectively change the semantic information transferred for the poisoned input samples to a target meaning. As the performance of semantic communications improves with the signal-to-noise ratio and the number of channel uses, the success of the backdoor attack increases as well. Also, increasing the Trojan ratio in training data makes the attack more successful. In the meantime, the effect of this attack on the unpoisoned input samples remains limited. Overall, this paper shows that the backdoor attack poses a serious threat to semantic communications and presents novel design guidelines to preserve the meaning of transferred information in the presence of backdoor attacks.

Semantic communications seeks to transfer information from a source while conveying a desired meaning to its destination. We model the transmitter-receiver functionalities as an autoencoder followed by a task classifier that evaluates the meaning of the information conveyed to the receiver. The autoencoder consists of an encoder at the transmitter to jointly model source coding, channel coding, and modulation, and a decoder at the receiver to jointly model demodulation, channel decoding and source decoding. By augmenting the reconstruction loss with a semantic loss, the two deep neural networks (DNNs) of this encoder-decoder pair are interactively trained with the DNN of the semantic task classifier. This approach effectively captures the latent feature space and reliably transfers compressed feature vectors with a small number of channel uses while keeping the semantic loss low. We identify the multi-domain security vulnerabilities of using the DNNs for semantic communications. Based on adversarial machine learning, we introduce test-time (targeted and non-targeted) adversarial attacks on the DNNs by manipulating their inputs at different stages of semantic communications. As a computer vision attack, small perturbations are injected to the images at the input of the transmitter's encoder. As a wireless attack, small perturbations signals are transmitted to interfere with the input of the receiver's decoder. By launching these stealth attacks individually or more effectively in a combined form as a multi-domain attack, we show that it is possible to change the semantics of the transferred information even when the reconstruction loss remains low. These multi-domain adversarial attacks pose as a serious threat to the semantics of information transfer (with larger impact than conventional jamming) and raise the need of defense methods for the safe adoption of semantic communications.

Communications systems to date are primarily designed with the goal of reliable (error-free) transfer of digital sequences (bits). Next generation (NextG) communication systems are beginning to explore shifting this design paradigm of reliably decoding bits to reliably executing a given task. Task-oriented communications system design is likely to find impactful applications, for example, considering the relative importance of messages. In this paper, a wireless signal classification is considered as the task to be performed in the NextG Radio Access Network (RAN) for signal intelligence and spectrum awareness applications such as user equipment (UE) identification and authentication, and incumbent signal detection for spectrum co-existence. For that purpose, edge devices collect wireless signals and communicate with the NextG base station (gNodeB) that needs to know the signal class. Edge devices may not have sufficient processing power and may not be trusted to perform the signal classification task, whereas the transfer of the captured signals from the edge devices to the gNodeB may not be efficient or even feasible subject to stringent delay, rate, and energy restrictions. We present a task-oriented communications approach, where all the transmitter, receiver and classifier functionalities are jointly trained as two deep neural networks (DNNs), one for the edge device and another for the gNodeB. We show that this approach achieves better accuracy with smaller DNNs compared to the baselines that treat communications and signal classification as two separate tasks. Finally, we discuss how adversarial machine learning poses a major security threat for the use of DNNs for task-oriented communications. We demonstrate the major performance loss under backdoor (Trojan) attacks and adversarial (evasion) attacks that target the training and test processes of task-oriented communications.

We propose RANA, a relightable and articulated neural avatar for the photorealistic synthesis of humans under arbitrary viewpoints, body poses, and lighting. We only require a short video clip of the person to create the avatar and assume no knowledge about the lighting environment. We present a novel framework to model humans while disentangling their geometry, texture, and also lighting environment from monocular RGB videos. To simplify this otherwise ill-posed task we first estimate the coarse geometry and texture of the person via SMPL+D model fitting and then learn an articulated neural representation for photorealistic image generation. RANA first generates the normal and albedo maps of the person in any given target body pose and then uses spherical harmonics lighting to generate the shaded image in the target lighting environment. We also propose to pretrain RANA using synthetic images and demonstrate that it leads to better disentanglement between geometry and texture while also improving robustness to novel body poses. Finally, we also present a new photorealistic synthetic dataset, Relighting Humans, to quantitatively evaluate the performance of the proposed approach.