Buying Warner Bros. Gives Netflix What It's Always Needed: An Identity The $83 billion deal gives the streamer a century's worth of prestige television and movies, from Batman movies to . It also ends the streaming wars. In a deal to acquire Warner Bros. announced Friday, Netflix will be scooping up HBO's many titles, including Courtesy of HBO Close your eyes, think for a minute, and tell me: What is a Netflix Movie? OK, try again: What is a Netflix Show? Sure, it's easy to rattle off some killer titles--, --but Netflix has never really had a brand identity.

A widespread folklore for explaining the success of Convolutional Neural Networks (CNNs) is that CNNs use a more compact representation than the Fully-connected Neural Network (FNN) and thus require fewer training samples to accurately estimate their parameters. We initiate the study of rigorously characterizing the sample complexity of estimating CNNs. We show that for an $m$-dimensional convolutional filter with linear activation acting on a $d$-dimensional input, the sample complexity of achieving population prediction error of $\epsilon$ is $\widetilde{O(m/\epsilon^2)$, whereas the sample-complexity for its FNN counterpart is lower bounded by $\Omega(d/\epsilon^2)$ samples. Since, in typical settings $m \ll d$, this result demonstrates the advantage of using a CNN. We further consider the sample complexity of estimating a one-hidden-layer CNN with linear activation where both the $m$-dimensional convolutional filter and the $r$-dimensional output weights are unknown. For this model, we show that the sample complexity is $\widetilde{O}\left((m+r)/\epsilon^2\right)$ when the ratio between the stride size and the filter size is a constant. For both models, we also present lower bounds showing our sample complexities are tight up to logarithmic factors. Our main tools for deriving these results are a localized empirical process analysis and a new lemma characterizing the convolutional structure. We believe that these tools may inspire further developments in understanding CNNs.

Transformers, with their attention mechanisms, have emerged as the state-of-the-art architectures of sequential modeling and empirically outperform feed-forward neural networks (FFNNs) across many fields, such as natural language processing and computer vision. However, their generalization ability, particularly for low-sensitivity functions, remains less studied. We bridge this gap by analyzing transformers on the $k$-parity problem. Daniely and Malach (NeurIPS 2020) show that FFNNs with one hidden layer and $O(nk^7 \log k)$ parameters can learn $k$-parity, where the input length $n$ is typically much larger than $k$. In this paper, we prove that FFNNs require at least $\Omega(n)$ parameters to learn $k$-parity, while transformers require only $O(k)$ parameters, surpassing the theoretical lower bound needed by FFNNs. We further prove that this parameter efficiency cannot be achieved with fixed attention heads. Our work establishes transformers as theoretically superior to FFNNs in learning parity function, showing how their attention mechanisms enable parameter-efficient generalization in functions with low sensitivity.

A widespread folklore for explaining the success of Convolutional Neural Networks (CNNs) is that CNNs use a more compact representation than the Fully-connected Neural Network (FNN) and thus require fewer training samples to accurately estimate their parameters. We initiate the study of rigorously characterizing the sample complexity of estimating CNNs. We show that for an m -dimensional convolutional filter with linear activation acting on a d -dimensional input, the sample complexity of achieving population prediction error of \epsilon is \widetilde{O(m/\epsilon 2), whereas the sample-complexity for its FNN counterpart is lower bounded by \Omega(d/\epsilon 2) samples. Since, in typical settings m \ll d, this result demonstrates the advantage of using a CNN. We further consider the sample complexity of estimating a one-hidden-layer CNN with linear activation where both the m -dimensional convolutional filter and the r -dimensional output weights are unknown.

The authors consider the number of samples needed to achieve an error epsilon in the context of learning an m-dimensional convolutional filter as well as one followed by a linear projection. This is motivated by a desire to rigorously understand the empirical success of CNNs. This paper seems technically correct, yet I believe the setting is very far from real CNNs to the point where it's not clear if the results will be impactful. The authors only consider a linear convolution layer, which corresponds to a wiener filtering-like operation according to their model, for removing noise for estimating the label. My concern is the motivation, the novelty and the assumptions.

Business leaders and workers have been talking for a long time about how artificial intelligence (AI) might change the future of work. More recently, the development of more sophisticated AI tools has led to a heightened level of discussion on the topic. In addition, ground-breaking advancements--as we see from the rapid rise of ChatGPT--have brought the conversation mainstream. I sat down with Kate O'Neill--best-selling author and founder of KO Insights--who possesses an interesting perspective not widely held by senior leaders and executives regarding AI. Known as the "Tech Humanist," Kate believes that AI should optimize the human experience, not replace it. She believes businesses need to think beyond how technology can help them meet their business goals.

With the development of natural language processing techniques(NLP), automatic diagnosis of eye diseases using ophthalmology electronic medical records (OEMR) has become possible. It aims to evaluate the condition of both eyes of a patient respectively, and we formulate it as a particular multi-label classification task in this paper. Although there are a few related studies in other diseases, automatic diagnosis of eye diseases exhibits unique characteristics. First, descriptions of both eyes are mixed up in OEMR documents, with both free text and templated asymptomatic descriptions, resulting in sparsity and clutter of information. Second, OEMR documents contain multiple parts of descriptions and have long document lengths. Third, it is critical to provide explainability to the disease diagnosis model. To overcome those challenges, we present an effective automatic eye disease diagnosis framework, NEEDED. In this framework, a preprocessing module is integrated to improve the density and quality of information. Then, we design a hierarchical transformer structure for learning the contextualized representations of each sentence in the OEMR document. For the diagnosis part, we propose an attention-based predictor that enables traceable diagnosis by obtaining disease-specific information. Experiments on the real dataset and comparison with several baseline models show the advantage and explainability of our framework.

Despite all the panicky warnings seen in the mainstream media, AI will not be taking over and automating peoples' jobs. AI will be replacing manual tasks, not job categories. However, something very important is missing from the picture: the involvement of the employees who will be charged with making AI and data-driven enterprises work. AI is only ramping up demand for the human talent needed to guide AI systems to engage in tasks relevant to the business, monitor and maintain the fairness and actionability of AI decisions, and to build, program, update, and ultimately retire these systems. That's one of the takeaways of Deloitte's latest research on the state of AI, which finds a lack of employee input into the ways AI will be deployed and what it will deliver.

Engineering has a white-male problem. Women make up just 14.5 percent of the engineering workforce in the United Kingdom, with ethnic minorities constituting just 8 percent. For Lila Ibrahim, chief operating officer at DeepMind, and Hayaatun Sillem, CEO of the Royal Academy of Engineering, being both female and people of color meant the odds were stacked against them in their industry. But for Sillem, who is the first woman and ethnic minority to hold her position, coming from such a diverse background helped her "to build empathy into her life"--a trait she describes as a superpower. And as for Ibrahim, the daughter of immigrants to the United States, she always felt like the "oddball" growing up in midwestern America.

Artificial intelligence would strengthen intelligence within companies, and would do the same for IT workshops. For example, AIOps (artificial intelligence for IT operations) applies AI and machine learning to data from IT processes, sifting through noise to detect, highlight and prevent problems. AI and machine learning also find their place in another emerging area of IT: helping DevOps teams ensure the viability and quality of software that moves at ever-increasing speeds through the system and to users. . As a recent survey by GitHub indicates, development and operations teams are massively turning to AI to streamline code flow in the software review and testing phase. The survey also reveals that 37% of teams are using AI/ML in software testing (up from 25% previously), and another 20% plan to use it this year.