Over-parameterized deep neural networks (DNNs) with sufficient capacity to memorize random noise can achieve excellent generalization performance, challenging the bias-variance trade-off in classical learning theory. Recent studies claimed that DNNs first learn simple patterns and then memorize noise; some other works showed a phenomenon that DNNs have a spectral bias to learn target functions from low to high frequencies during training. However, we show that the monotonicity of the learning bias does not always hold: under the experimental setup of deep double descent, the high-frequency components of DNNs diminish in the late stage of training, leading to the second descent of the test error. Besides, we find that the spectrum of DNNs can be applied to indicating the second descent of the test error, even though it is calculated from the training set only.

Right off the bat, here is a scene from the near-future: AI takes a look at John Jones' medical records, does instant collating, and comes up with a disease diagnosis. Via Zoom, the doctor's AI assistant slaps on a diagnosis, and an hour later, two bottles of medical drugs arrive at Jones' door. One problem: the data set assembled by AI is preposterous. Jones' so-called symptoms don't add up to a disease. Only in another data set, held by the CDC, do the symptoms require a disease-label.

The Regent Honeyeaters of Australasia are forgetting how to talk. The songbird's habitat has been so severely devastated that its numbers are dwindling. Worse, the ones that remain are so scattered that the adult males are too far apart to teach the young how to sing for a mate -- how to speak their own language. The gradual loss of the Honeyeaters' song, their primary tool for wooing a partner, creates a vicious circle of spiraling decline. Humans, on the other hand, cannot shut up.

AI Policy Matters is a regular column in the ACM SIGAI AI Matters newsletter featuring summaries and commentary based on postings that appear twice a month in the AI Matters blog. Facial recognition (FR) issues continue to appear in the news, as well as in scholarly journal articles, while FR systems are being banned and some research is shown to be bad science. AI system researchers who try to associate facial technology output with human characteristics are sometimes referred to as machine-assisted phrenologists. Problems with FR research have been demonstrated in machine learning research such as work by Steed and Caliskan in "A set of distinct facial traits learned by machines is not predictive of appearance bias in the wild." Meanwhile many examples of harmful products and misuses have been identified in areas such as criminality, video interviewing, and many others. Some communities have considered bans.

A composition of 10,000 unique cat images generated by the artificial intelligence called GAN (generative adversarial network), no human was involved in the creation of these cats, a machine created them by machine learning and artificial intelligence algorithms. The owner will receive the full resolution at 10,000 x 10,000 pixels (100 megapixels) and can ask me to send the full resolution cats images.

A collage of 10,000 unique nonexistent persons generated by the artificial intelligence called GAN (generative adversarial network), no human was involved in the creation of these person images, a machine created them by machine learning and artificial intelligence algorithms. The owner will receive the full resolution at 10,000 x 10,000 pixels (100 megapixels) and can ask me to send the full resolution person images.

Do you find artificial intelligence and its capabilities fascinating? Then you'll want to check out these sci-fi novels with stories about robots and AI bots. The novels in this list include anti-hero tales about robot uprising and stories about AI bots co-existing with humans and falling in love with them too. Sci-fi AI novels offer an innovative and imaginative insight into this disruptive technology that we see around in our day-to-day lives. Every recommendation in this list is unique, which raises many questions in curious minds.

The United Nations should follow the European Commission in establishing a regulatory framework for artificial intelligence and biometrics to protect people subject to the technologies, build trust in their use and take the pressure off data scientists to constantly justify the ethics, writes Eleonore Fournier-Tombs of McGill University for The Conversation. The European Commission (EC) put forward proposals in April 2021 that seek to harmonize rules on artificial intelligence and create mechanisms which Fournier-Tombs likens to the process for seeking approval for a new drug. Developers of a new high-risk application of AI would have to submit it for regulatory approval. They would also have to provide details on how the models and data are used and how impacts on privacy or discrimination would be addressed. Areas of risk include biometric identification, categorization and evaluation of the eligibility of people for accessing welfare and services, including in emergency response situations.

Explainable AI (XAI) fully describes an AI model, its expected impact and any potential biases. It helps you understand the steps taken by an AI technique to arrive at a decision. In this article, we will take a look at XAI in detail and explore how you can implement it in your organisation. "About half (46%) of South African companies indicate that they are already implementing AI within their organisations." Why is explainable AI important for your business?

Lossy compression algorithms are typically designed to achieve the lowest possible distortion at a given bit rate. However, recent studies show that pursuing high perceptual quality would lead to increase of the lowest achievable distortion (e.g., MSE). This paper provides nontrivial results theoretically revealing that, \textit{1}) the cost of achieving perfect perception quality is exactly a doubling of the lowest achievable MSE distortion, \textit{2}) an optimal encoder for the "classic" rate-distortion problem is also optimal for the perceptual compression problem, \textit{3}) distortion loss is unnecessary for training a perceptual decoder. Further, we propose a novel training framework to achieve the lowest MSE distortion under perfect perception constraint at a given bit rate. This framework uses a GAN with discriminator conditioned on an MSE-optimized encoder, which is superior over the traditional framework using distortion plus adversarial loss. Experiments are provided to verify the theoretical finding and demonstrate the superiority of the proposed training framework.