Self-supervised learning

Self-supervised learning (SSL) is a paradigm in machine learning where a model is trained on a task using the data itself to generate supervisory signals, rather than relying on externally-provided labels. In the context of neural networks, self-supervised learning aims to leverage inherent structures or relationships within the input data to create meaningful training signals. SSL tasks are designed so that solving them requires capturing essential features or relationships in the data. The input data is typically augmented or transformed in a way that creates pairs of related samples, where one sample serves as the input, and the other is used to formulate the supervisory signal. This augmentation can involve introducing noise, cropping, rotation, or other transformations. Self-supervised learning more closely imitates the way humans learn to classify objects. During SSL, the model learns in two steps. First, the task is solved based on an auxiliary or pretext classification task using pseudo-labels, which help to initialize the model parameters. Next, the actual task is performed with supervised or unsupervised learning. Self-supervised learning has produced promising results in recent years, and has found practical application in fields such as audio processing, and is being used by Facebook and others for speech recognition. == Pseudo-labels == Pseudo-labels are automatically generated labels that a model assigns to unlabeled data based on its own predictions. They are widely used in self-supervised and semi-supervised learning, where ground-truth annotations are limited or unavailable. By treating predicted labels as surrogate ground truth, learning algorithms can make use of large quantities of unlabeled data in the training process. Pseudo-labeling also plays an important role in systems that must adapt to concept drift, where the statistical properties of the data change over time. In these scenarios, the model may detect that an incoming instance deviates from previously learned behavior. The system then generates a classification result for that instance, and this predicted class is used as a pseudo-label for updating or retraining model components that are becoming outdated. This approach enables continuous adaptation in dynamic environments without requiring manual annotation. In many adaptive learning pipelines, pseudo-labels are chosen when the classifier produces sufficiently confident predictions, reducing the risk of propagating errors. These pseudo-labeled instances are then incorporated into training to refresh or evolve the model's understanding of emerging data patterns, particularly when existing components show signs of “aging” due to drift or distributional shifts. This strategy reduces reliance on manual labeling while helping maintain long-term model performance. == Types == === Autoassociative self-supervised learning === Autoassociative self-supervised learning is a specific category of self-supervised learning where a neural network is trained to reproduce or reconstruct its own input data. In other words, the model is tasked with learning a representation of the data that captures its essential features or structure, allowing it to regenerate the original input. The term "autoassociative" comes from the fact that the model is essentially associating the input data with itself. This is often achieved using autoencoders, which are a type of neural network architecture used for representation learning. Autoencoders consist of an encoder network that maps the input data to a lower-dimensional representation (latent space), and a decoder network that reconstructs the input from this representation. The training process involves presenting the model with input data and requiring it to reconstruct the same data as closely as possible. The loss function used during training typically penalizes the difference between the original input and the reconstructed output (e.g. mean squared error). By minimizing this reconstruction error, the autoencoder learns a meaningful representation of the data in its latent space. === Contrastive self-supervised learning === For a binary classification task, training data can be divided into positive examples and negative examples. Positive examples are those that match the target. For example, if training a classifier to identify birds, the positive training data would include images that contain birds. Negative examples would be images that do not. Contrastive self-supervised learning uses both positive and negative examples. The loss function in contrastive learning is used to minimize the distance between positive sample pairs, while maximizing the distance between negative sample pairs. An early example uses a pair of 1-dimensional convolutional neural networks to process a pair of images and maximize their agreement. Contrastive Language-Image Pre-training (CLIP) allows joint pretraining of a text encoder and an image encoder, such that a matching image-text pair have image encoding vector and text encoding vector that span a small angle (having a large cosine similarity). InfoNCE (Noise-Contrastive Estimation) is a method to optimize two models jointly, based on Noise Contrastive Estimation (NCE). Given a set X = { x 1 , … x N } {\displaystyle X=\left\{x_{1},\ldots x_{N}\right\}} of N {\displaystyle N} random samples containing one positive sample from p ( x t + k ∣ c t ) {\displaystyle p\left(x_{t+k}\mid c_{t}\right)} and N − 1 {\displaystyle N-1} negative samples from the 'proposal' distribution p ( x t + k ) {\displaystyle p\left(x_{t+k}\right)} , it minimizes the following loss function: L N = − E X [ log ⁡ f k ( x t + k , c t ) ∑ x j ∈ X f k ( x j , c t ) ] {\displaystyle {\mathcal {L}}_{\mathrm {N} }=-\mathbb {E} _{X}\left[\log {\frac {f_{k}\left(x_{t+k},c_{t}\right)}{\sum _{x_{j}\in X}f_{k}\left(x_{j},c_{t}\right)}}\right]} === Non-contrastive self-supervised learning === Non-contrastive self-supervised learning (NCSSL) uses only positive examples. Counterintuitively, NCSSL converges on a useful local minimum rather than reaching a trivial solution, with zero loss. For the example of binary classification, it would trivially learn to classify each example as positive. Effective NCSSL requires an extra predictor on the online side that does not back-propagate on the target side. === Joint-Embedding and Predictive Architectures === A major class of self-supervised learning moves beyond contrastive pairs, instead maximizing the agreement between views while preventing collapse through statistical constraints. Rooted in Deep Canonical Correlation Analysis (Deep CCA), this approach includes Joint-Embedding Architectures (JEA) like Barlow Twins and VICReg, which enforce covariance constraints to learn invariant representations without negative sampling. Deep Latent Variable Path Modelling (DLVPM) generalizes this to multimodal systems, using path models to enforce correlation and orthogonality across diverse data types. In 2022 Yann LeCun introduced Joint-Embedding Predictive Architectures (JEPA) as a step towards decision making, reasoning, and autonomous human intelligence in machines, including self-improvement through autonomous learning. Founded in representation learning, LeCun included the concept of a “world model” in JEPA which aims to enable machines to replicate human intellect by providing machines with a concept for the world in which they exist. Unlike autoencoders, JEPAs operate entirely in latent space, avoiding pixel-level noise to focus on semantic structure. Rather than just learning invariance, JEPAs learn by predicting masked latent representations from visible context. JEPA has been applied to domains such as image analysis, audio processing, and motion in images and video. == Comparison with other forms of machine learning == SSL belongs to supervised learning methods insofar as the goal is to generate a classified output from the input. At the same time, however, it does not require the explicit use of labeled input-output pairs. Instead, correlations, metadata embedded in the data, or domain knowledge present in the input are implicitly and autonomously extracted from the data. These supervisory signals, extracted from the data, can then be used for training. SSL is similar to unsupervised learning in that it does not require labels in the sample data. Unlike unsupervised learning, however, learning is not done using inherent data structures. Semi-supervised learning combines supervised and unsupervised learning, requiring only a small portion of the learning data be labeled. In transfer learning, a model designed for one task is reused on a different task. Training an autoencoder intrinsically constitutes a self-supervised process, because the output pattern needs to become an optimal reconstruction of the input pattern itself. However, in current jargon, the term 'self-supervised' often refers to tasks based on a pretext-task training setup

Face Swap Live

Face Swap Live is a mobile app created by Laan Labs that enables users to swap faces with another person in real-time using the device's camera. It was released on December 14, 2015. In addition to swapping faces with another person, the app enables users to create videos using a set of bundled live filters. The app is available on iOS and Android devices. Face Swap Live was named Apple's #2 best-selling paid app in 2016.

Initialization vector

In cryptography, an initialization vector (IV) or starting variable is an input to a cryptographic primitive being used to provide the initial state. The IV is typically required to be random or pseudorandom, but sometimes an IV only needs to be unpredictable or unique. Randomization is crucial for some encryption schemes to achieve semantic security, a property whereby repeated usage of the scheme under the same key does not allow an attacker to infer relationships between (potentially similar) segments of the encrypted message. For block ciphers, the use of an IV is described by the modes of operation. Some cryptographic primitives require the IV only to be non-repeating, and the required randomness is derived internally. In this case, the IV is commonly called a nonce (a number used only once), and the primitives (e.g. CBC) are considered stateful rather than randomized. This is because an IV need not be explicitly forwarded to a recipient but may be derived from a common state updated at both sender and receiver side. (In practice, a short nonce is still transmitted along with the message to consider message loss.) An example of stateful encryption schemes is the counter mode of operation, which has a sequence number for a nonce. The IV size depends on the cryptographic primitive used; for block ciphers it is generally the cipher's block-size. In encryption schemes, the unpredictable part of the IV has at best the same size as the key to compensate for time/memory/data tradeoff attacks. When the IV is chosen at random, the probability of collisions due to the birthday problem must be taken into account. Traditional stream ciphers such as RC4 do not support an explicit IV as input, and a custom solution for incorporating an IV into the cipher's key or internal state is needed. Some designs realized in practice are known to be insecure; the WEP protocol is a notable example, and is prone to related-IV attacks. == Motivation == A block cipher is one of the most basic primitives in cryptography, and frequently used for data encryption. However, by itself, it can only be used to encode a data block of a predefined size, called the block size. For example, a single invocation of the AES algorithm transforms a 128-bit plaintext block into a ciphertext block of 128 bits in size. The key, which is given as one input to the cipher, defines the mapping between plaintext and ciphertext. If data of arbitrary length is to be encrypted, a simple strategy is to split the data into blocks each matching the cipher's block size, and encrypt each block separately using the same key. This method is not secure as equal plaintext blocks get transformed into equal ciphertexts, and a third party observing the encrypted data may easily determine its content even when not knowing the encryption key. To hide patterns in encrypted data while avoiding the re-issuing of a new key after each block cipher invocation, a method is needed to randomize the input data. In 1980, the NIST published a national standard document designated Federal Information Processing Standard (FIPS) PUB 81, which specified four so-called block cipher modes of operation, each describing a different solution for encrypting a set of input blocks. The first mode implements the simple strategy described above, and was specified as the electronic codebook (ECB) mode. In contrast, each of the other modes describe a process where ciphertext from one block encryption step gets intermixed with the data from the next encryption step. To initiate this process, an additional input value is required to be mixed with the first block, and which is referred to as an initialization vector. For example, the cipher-block chaining (CBC) mode requires an unpredictable value, of size equal to the cipher's block size, as additional input. This unpredictable value is added to the first plaintext block before subsequent encryption. In turn, the ciphertext produced in the first encryption step is added to the second plaintext block, and so on. The ultimate goal for encryption schemes is to provide semantic security: by this property, it is practically impossible for an attacker to draw any knowledge from observed ciphertext. It can be shown that each of the three additional modes specified by the NIST are semantically secure under so-called chosen-plaintext attacks. == Properties == Properties of an IV depend on the cryptographic scheme used. A basic requirement is uniqueness, which means that no IV may be reused under the same key. For block ciphers, repeated IV values devolve the encryption scheme into electronic codebook mode: equal IV and equal plaintext result in equal ciphertext. In stream cipher encryption uniqueness is crucially important as plaintext may be trivially recovered otherwise. Example: Stream ciphers encrypt plaintext P to ciphertext C by deriving a key stream K from a given key and IV and computing C as C = P xor K. Assume that an attacker has observed two messages C1 and C2 both encrypted with the same key and IV. Then knowledge of either P1 or P2 reveals the other plaintext since C1 xor C2 = (P1 xor K) xor (P2 xor K) = P1 xor P2. Many schemes require the IV to be unpredictable by an adversary. This is effected by selecting the IV at random or pseudo-randomly. In such schemes, the chance of a duplicate IV is negligible, but the effect of the birthday problem must be considered. As for the uniqueness requirement, a predictable IV may allow recovery of (partial) plaintext. Example: Consider a scenario where a legitimate party called Alice encrypts messages using the cipher-block chaining mode. Consider further that there is an adversary called Eve that can observe these encryptions and is able to forward plaintext messages to Alice for encryption (in other words, Eve is capable of a chosen-plaintext attack). Now assume that Alice has sent a message consisting of an initialization vector IV1 and starting with a ciphertext block CAlice. Let further PAlice denote the first plaintext block of Alice's message, let E denote encryption, and let PEve be Eve's guess for the first plaintext block. Now, if Eve can determine the initialization vector IV2 of the next message she will be able to test her guess by forwarding a plaintext message to Alice starting with (IV2 xor IV1 xor PEve); if her guess was correct this plaintext block will get encrypted to CAlice by Alice. This is because of the following simple observation: CAlice = E(IV1 xor PAlice) = E(IV2 xor (IV2 xor IV1 xor PAlice)). Depending on whether the IV for a cryptographic scheme must be random or only unique the scheme is either called randomized or stateful. While randomized schemes always require the IV chosen by a sender to be forwarded to receivers, stateful schemes allow sender and receiver to share a common IV state, which is updated in a predefined way at both sides. == Block ciphers == Block cipher processing of data is usually described as a mode of operation. Modes are primarily defined for encryption as well as authentication, though newer designs exist that combine both security solutions in so-called authenticated encryption modes. While encryption and authenticated encryption modes usually take an IV matching the cipher's block size, authentication modes are commonly realized as deterministic algorithms, and the IV is set to zero or some other fixed value. == Stream ciphers == In stream ciphers, IVs are loaded into the keyed internal secret state of the cipher, after which a number of cipher rounds are executed prior to releasing the first bit of output. For performance reasons, designers of stream ciphers try to keep that number of rounds as small as possible, but because determining the minimal secure number of rounds for stream ciphers is not a trivial task, and considering other issues such as entropy loss, unique to each cipher construction, related-IVs and other IV-related attacks are a known security issue for stream ciphers, which makes IV loading in stream ciphers a serious concern and a subject of ongoing research. == WEP IV == The 802.11 encryption algorithm called WEP (short for Wired Equivalent Privacy) used a short, 24-bit IV, leading to reused IVs with the same key, which led to it being easily cracked. Packet injection allowed for WEP to be cracked in times as short as several seconds. This ultimately led to the deprecation of WEP. == SSL 2.0 IV == In cipher-block chaining mode (CBC mode), the IV need not be secret, but must be unpredictable (In particular, for any given plaintext, it must not be possible to predict the IV that will be associated to the plaintext in advance of the generation of the IV.) at encryption time. Additionally for the output feedback mode (OFB mode), the IV must be unique. In particular, the (previously) common practice of re-using the last ciphertext block of a message as the IV for the next message is insecure (for example, this method was used by SSL 2.0). If an attacker knows

Social media surgery

A social media surgery is a gathering at which volunteer "surgeons" with expertise in using web tools, chiefly social media, offer free advice in using such tools, to representatives ("patients") of non-profit organisations, charities, community groups and activists, with "no boring speeches or jargon". The idea was conceived by Pete Ashton, with Nick Booth of Podnosh Ltd, who ran the first such surgery in Birmingham, England, on 15 October 2008. In July 2009, a spin-off surgery (dubbed the "Social media mob") started in Mosman, Australia, and in January 2010, the first spin-off surgery in Africa was held. On 16 February 2012, it was announced that the Social Media Surgery movement had won "the Prime Minister’s Big Society Award". Prime Minister David Cameron said: This is an excellent initiative - such a simple idea and yet so effective. The popularity of these surgeries and the fact that they have inspired so many others across the country to follow in their footsteps, is testament to its brilliance. Congratulations to Nick and all the volunteers who have shared their time and expertise to help so many local groups make the most of the internet to support their community. A great example of the Big Society in action. The scheme also won the 2013 Adult Learners' Week "BBC Learning Through Technology Award".

Social media reach

Social media reach is a media analytics metric that refers to the number of users who have come across a particular content on a particular social media platform. Social media platforms have their own individual ways of tracking, analyzing and reporting the traffic on each of the individual platforms. As these platforms are a main source of communication between companies and their target audiences, by conducting research, companies are able to utilize analytical information, such as the reach of their posts, to better understand the interactions between the users and their content. There are multiple underlying factors that will determine what shows up on a newsfeed or timeline. Algorithms, for example, are a type of factor that can alter the reach of a post due to the way the algorithm is coded, which can affect who sees a post and when. Other examples of factors that can impede the reach can include the time at which posts are made, as well as how frequent the posts are between one another. In comparison, an impression is the total number of circumstances where content has been shown on a social timeline, meanwhile, engagement looks at how people interact with the content that they see on a social platform such as like, share or retweet. == Reach on Facebook == Facebook has their own analytic platform which allows the user to see how other users are interacting with their posts, with the use of multiple metrics. This is not something the average user uses, but rather a tool that is used by pages or public figures. For example, Facebook pages that represent a business often look at the activity their posts have generated. There are three types of reach that can be looked at on the Facebook analytic platform. === Types of reach === ==== Organic Reach ==== This type of reach regards the number of distinct users that have seen a specific post on their feed. Organic reach, in other words is the number of people who have seen the post being analyzed on their Facebook newsfeed. Data gathered from this type of reach can give intel to those doing the analysis, such as the demographics of those who have seen the post. ==== Paid Reach ==== This type of reach regards the number of times that distinct users have come across sponsored posts, ads or content. In other words, paid reach is the number of times Facebook users have seen a post that has been paid for by a company. Data collected can give insight, to advertisers or marketers for example, on the activity based around the reach of their post. ==== Viral Reach ==== This type of reach regards the number of views by distinct users on posts that have been commented on or shared by their friends on Facebook. In other words, viral reach looks at the number of people who have seen a post after a friend of theirs commented or shared the original post, therefore it showed on their timeline. Viral reach can be looked at in terms of a collective number of times that the post has been on individual user's timelines. Data collected from viral reach can be used in multiple ways, for example, it can be used to analyze the type of content that gets shared or commented on and can be further used to compare to other posts. === Engaged users === This refers to the number of individual users who have clicked and interacted with a post on Facebook. == Reach on Twitter == Twitter gives access to any of their users to analytics of their tweets as well as their followers. Their dashboard is user friendly, which allows anyone to take a look at the analytics behind their Twitter account. This open access is useful for both the average user and companies as it can provide a quick glance or general outlook of who has seen their tweets. The way that Twitter works is slightly different than the way of Facebook in terms of the reach. On Twitter, especially for users with a higher profile, they are not only engaging with the people who follow them, but also with the followers of their own followers. The reach metric on Twitter looks at the quantity of Twitter users who have been engaged, but also the number of users that follow them as well. This metric is useful to see the if the tweets/content being shared on Twitter are contributing to the growth of audience on this platform. == Reach on Instagram == Instagram gives their users access to their reach, in the Instagram Insights section. Instagram insights can be used to learn more about an account's followers and performance. Reach indicates the total number of unique Instagram accounts that have seen your Instagram post or story. You can find this data by looking at each individual post insights. == Uses of reach == The reach can be a useful metric to analyze for marketers and advertisers. Social media is a platform that is used by marketers to directly target their intended audience with ease. These platforms not only allow marketers to get a better understanding of their audience, but also allow advertisers to insert their ads onto the timelines of specific users to later be able to conduct research to see the reach of their posts/content. The basic goal of marketers is to increase their reach as much as possible to impact bigger audiences of their dream customers and, in the end, make more sales. When doing organic social media marketing, using paid methods like ads or doing influencer marketing whether it is paid or free, it allows marketers to track the performance of their strategy and tweak it based on what works and what does not. == Analytics and reach == Social analytics looks at the data collected based on the interactions of users on social media platforms. A lot of information can be gathered which can provide intel based on user activities on social media. When looking into analytics in regard to social media, each company or group has a different goal in mind to engage their audience. At a glance, the three might seem as if they are very similar, however the differences between them are significant. There are many aspects that can be analyzed from the data gathered from social media platforms, depending on what is being observed, the correct metric would then be selected to further analyze. One example of the many metrics that can be used through social analytics is the reach. == Reach formula == To calculate social media reach one can use the following formula: R = I f ¯ {\displaystyle R={\frac {I}{\bar {f}}}} where R {\displaystyle R} — is social media reach, I {\displaystyle I} stands for the number of impressions, f ¯ {\displaystyle {\bar {f}}} is the average frequency of impressions per user. f ¯ {\displaystyle {\bar {f}}} represents the number of events when the ad is shown to a particular user. The average value should be calculated over the time period with stable settings of advertisement campaign. == Commenting For Better Reach == Commenting For Better Reach also known as "CFBR" is a widely used strategy for organically boosting post reach on social media platforms. Algorithms tend to favor posts with substantial likes and comments, granting them broader exposure compared to less engaging content. Primarily seen on LinkedIn, a platform geared toward professional networking and business connections, the use of CFBR signals active engagement aimed at enhancing post visibility. It is important to note that genuine and meaningful comments are key to effective engagement. Spammy or irrelevant comments not only detract from the conversation but may also limit a post's potential reach and impact.

Outline of deep learning

The following outline is provided as an overview of, and topical guide to, deep learning: Deep learning is a subfield of machine learning and artificial intelligence based on artificial neural networks with multiple processing layers. It emphasizes representation learning and is widely used in areas such as computer vision, natural language processing, speech recognition, recommender systems, robotics, and generative artificial intelligence. == Ways to categorize deep learning == A field of study A branch of artificial intelligence A subfield of machine learning A subfield of computer science A form of representation learning A class of methods based on artificial neural networks An approach used in computational statistics == History == === Precursors === Cybernetics Perceptron Connectionism Neocognitron Backpropagation === Milestones === LeNet Long short-term memory Deep belief network AlexNet Sequence to sequence learning Generative adversarial network Residual neural network Transformer BERT Generative pre-trained transformer Diffusion model === Related histories === History of artificial intelligence History of machine learning Timeline of machine learning == Core concepts == == Learning settings == Supervised learning Unsupervised learning Self-supervised learning Semi-supervised learning Reinforcement learning Transfer learning Multitask learning Multimodal learning Online machine learning Continual learning == Common tasks == Image classification Object detection Image segmentation Automatic speech recognition Neural machine translation Question answering Automatic summarization Text-to-image model Protein structure prediction == Architectures == === Feedforward and convolutional architectures === Feedforward neural network Multilayer perceptron Convolutional neural network Radial basis function network Residual neural network U-Net === Recurrent and sequence architectures === Recurrent neural network Long short-term memory Gated recurrent unit Sequence to sequence learning Recursive neural network === Representation-learning architectures === Autoencoder Denoising autoencoder Sparse autoencoder Variational autoencoder Restricted Boltzmann machine Deep belief network === Attention and transformer architectures === Attention (machine learning) Transformer BERT Generative pre-trained transformer Vision transformer === Generative and probabilistic architectures === Autoregressive model Diffusion model Energy-based model Generative adversarial network Mixture of experts === Graph and memory architectures === Graph neural network Graph convolutional network Siamese network Neural Turing machine Memory network Echo state network Capsule neural network == Neural network components and techniques == Artificial neuron Activation function Rectified linear unit Sigmoid function Softmax function Embedding Convolution Pooling layer Attention Batch normalization Layer normalization Residual connections == Training and optimization == Backpropagation Gradient descent Stochastic gradient descent Adam optimization Learning rate Loss function Cross-entropy Mean squared error Regularization Dropout Early stopping Batch normalization Data augmentation Transfer learning Knowledge distillation Ensemble learning Curriculum learning == Datasets and benchmarks == CIFAR-10 ImageNet MNIST database Common Objects in Context (COCO) General Language Understanding Evaluation (GLUE) benchmark LibriSpeech SQuAD == Applications == === Computer vision === Computer vision Facial recognition system Image classification Image segmentation Medical imaging Object detection Optical character recognition === Natural language processing === Automatic summarization Chatbot Information retrieval Large language model Natural language processing Neural machine translation Question answering Sentiment analysis === Speech and audio === Automatic speech recognition Music information retrieval Speaker recognition Speech synthesis === Science and medicine === Bioinformatics Computational biology Drug discovery Medical diagnosis Protein structure prediction === Robotics and control === Autonomous car Computer game bot Control theory Robotics === Recommendation, search, and forecasting === Anomaly detection Forecasting Fraud detection Recommender system Search engine === Generative artificial intelligence === Deepfake Generative artificial intelligence Large language model Speech synthesis Text-to-image model === Computer graphics and video games === Deep Learning Anti-Aliasing (DLAA) Deep Learning Super Sampling (DLSS) == Hardware == AMD Instinct AMD XDNA Application-specific integrated circuit Deep learning processor, Neural processing unit (NPU), or Neural Engine Field-programmable gate array General-purpose computing on graphics processing units (GPGPU) Graphics processing unit NVIDIA Deep Learning Accelerator (NVDLA) Tensor processing unit Vision processing unit Wafer-scale integration === Supporting software platforms === CUDA Metal ROCm == Software == === Open-source frameworks and libraries === === Neural network software === EDLUT Emergent Encog JOONE Neuroph NeuroSolutions OpenNN Peltarion Synapse SNNS === Platforms, tools, and deployment === Amazon SageMaker Google Colab Hugging Face Kaggle Kubeflow MLflow ONNX OpenVINO TensorFlow Hub == Algorithms for deep learning and neural networks == Backpropagation Conjugate gradient method Generalized Hebbian algorithm Gradient descent Levenberg–Marquardt algorithm Perceptron Quasi-Newton method Wake-sleep algorithm == Methods and related topics == === Representation and metric learning === Contrastive learning Embedding Feature learning Manifold learning Metric learning === Generative modeling === Autoregressive model Diffusion model Generative adversarial network Generative model Variational inference === Efficient and scalable deep learning === Knowledge distillation Low-rank approximation Mixture of experts Quantization Sparsity === Reliability, safety, and interpretability === Adversarial machine learning AI alignment Algorithmic bias Catastrophic forgetting Differential privacy Explainable artificial intelligence Federated learning Hallucination (artificial intelligence) == Conferences and workshops == Annual Meeting of the Association for Computational Linguistics Conference on Computer Vision and Pattern Recognition Conference on Neural Information Processing Systems International Conference on Computer Vision International Conference on Learning Representations International Conference on Machine Learning == Organizations == === Research laboratories and institutions === Allen Institute for AI Alberta Machine Intelligence Institute European Laboratory for Learning and Intelligent Systems Google DeepMind Meta AI Mila Microsoft Research Vector Institute === Companies === Anthropic Cerebras Cohere DeepSeek Mistral AI OpenAI Stability AI xAI == Publications == === Books === Deep Learning – Ian Goodfellow and Yoshua Bengio Neural Networks and Deep Learning – Michael Nielsen Perceptrons – Marvin Minsky and Seymour Papert === Journals === IEEE Transactions on Neural Networks and Learning Systems Neural Networks Neural Computation == Influential persons ==

Online Safety Amendment (Social Media Minimum Age) Act 2024

The Online Safety Amendment (Social Media Minimum Age) Act 2024 is an Australian act of parliament that prohibits minors under the age of 16 from holding an account on certain social media platforms. It is an amendment to the Online Safety Act 2021 and was passed by the Parliament of Australia on 29 November 2024. It imposes monetary penalties on social media companies that fail to take reasonable steps to prevent minors under 16 that are located in Australia from having accounts on their services. The legislation allows the government to determine which social media platforms must ban age‑restricted users and proclaim a date for the commencement of the ban, with those provisions taking effect on 10 December 2025. Facebook, Instagram, Reddit, Snapchat, TikTok, Twitter, Threads, Twitch, Kick, and YouTube were age‑restricted on 10 December 2025, with the possibility that more platforms may be added. The act is being challenged in the High Court by the Digital Freedom Project. == Background == The ban on access to social media by young people by the federal government originated in November 2023, when shadow communications minister David Coleman introduced a private member's bill requiring the government to conduct a trial for age-verification technology on pornography and social media platforms. While the bill did not succeed, the Albanese government funded the trial in the 2024 Australian federal budget. In June 2024, opposition leader Peter Dutton pledged that a Coalition government would implement a ban on social media for under-16s within 100 days of taking office. The following month, prime minister Anthony Albanese announced the government would introduce legislation banning under-16s from social media. The Online Safety Amendment (Social Media Minimum Age) Bill 2024 was introduced into parliament by minister for communications Michelle Rowland on 21 November 2024, passing both houses on 28 November 2024. The ban on access to social media by young people by the federal government also gained momentum following an entreaty by the wife of the premier of South Australia, Peter Malinauskas, to her husband. She requested that he read The Anxious Generation by Jonathan Haidt and take action to address the impact of social media on the mental health of children. The couple have four young children, and, thinking of them, the premier thought that government should play a part in helping parents to regulate use of social media by their children at home. Malinauskas contacted former High Court chief justice Robert French, who agreed to look at the issue, and in September 2024 handed the premier a 267 page proposal, which he dubbed a "Swiss Army knife" rather than a machete, to adjust to social media's "changing landscape and its complexity". The leaders of other states and territories gave their support to Malinauskas's idea, and he took the French report to National Cabinet to collaborate with chief ministers, premiers, and the prime minister. Community support swelled after stories of parents who had lost their children to suicide after being bullied on social media were published. Albanese himself was moved by a personal letter received from Kelly O'Brien, whose 12-year-old daughter Charlotte had taken her own life due to bullying at school. An event took place at the sidelines of the United Nations General Assembly session in September 2025 at which a mother spoke of her daughter's suicide as "death by bullying ... enabled by social media". The speech won support from world leaders in Greece, Fiji, Tonga and the president of the European Commission Ursula von der Leyen. In early September 2024, South Australia proposed legislation similar to the federal law now in place. The state-based version was intended to ban users under the age of 14, unlike the federal law, which bans those under 16. The state-based law also proposed to require parental consent for 14 and 15‑year‑olds. Later in September, prime minister Anthony Albanese announced that his government intended to introduce legislation to set a minimum age requirement for social media. In November 2024, the federal government indicated their intention to engage the Age Check Certification Scheme following a tender process for an age assurance technology trial. The Albanese government's proposed ban was supported by the governments of every state and territory. Albanese described social media as a "scourge", and said "I want people to spend more time on the footy field or the netball court than they're spending on their phones", that family members are "worried sick about the safety of our kids online", and that social media "is having a negative impact on young people's mental health and on anxiety". Albanese's statements followed an earlier pledge by Liberal opposition leader Peter Dutton who was pushed by the early advocacy of shadow communications minister David Coleman to implement a ban on social media for under 16s within 100 days of being elected. The opposition organised an open letter signed by 140 experts who specialise in child welfare and technology. The opposition was concerned about the invasion of privacy that will occur with the introduction of identification-based age checks. An advocacy group for digital companies in Australia called the plans a "20th Century response to 21st Century challenges". A director of a mental health service voiced concerns, stating that "73% of young people across Australia who accessed mental health support did so through social media". == Implementation == Social media companies will receive a transition period of one year after the legislation is enacted to introduce reasonable controls preventing minors under the age of 16 from holding accounts on their services while physically located in Australia. Enforcement will involve fines of up to A$49.5 million for companies failing to take such steps, with no consequences for parents and children who violate the restrictions. There are no parental consent exceptions to the ban, and while the use of virtual private networks (VPNs) to access these services remains legal in Australia, the services are expected to try to stop under 16s from using VPNs to pretend to be outside Australia. The expectation is to make best-efforts to implement the ban on platforms including Facebook, Instagram, Reddit, Snapchat, TikTok, Twitter, Threads, Twitch, Kick and YouTube. Some social media companies are now obligated to become good enough at profiling Australian children under 16 to satisfy the Australian government they tried to implement the ban to avoid being fined. Consequently, social media companies said they will try to identify restricted users using various methods including behavioural inferencing. On 5 November 2025, it was announced that online gaming platform Roblox will not be banned, but Reddit and live-streaming platform Kick will be added to the list of platforms to be banned. A report by Age Check Certification Scheme, a UK company recruited by the government to consult on the technology used to implement the restrictions, was issued in June 2025, ahead of the December deadline to implement the ban. In June 2025, the preliminary report was released, which stated that "there are no significant technological barriers" to implementing the ban. In late July 2025, Google warned that it would sue the Australian government if YouTube was included in the ban. On 30 July, the government announced that it would extend its social media age limit to include YouTube, following advice from Grant. On 30 July 2025, the minister for communications, Anika Wells, published the Online Safety (Age-Restricted Social Media Platforms) Rules 2025, which specify exactly which types of social media platforms will be banned for certain users. On 31 August 2025, the full report was released, which stated that it would technically be possible to implement the ban; however, coordination among different services is required to successfully implement it. It also highlighted the benefits and flaws of different methods of age verification. On 16 September 2025, it was announced that the eSafety Commissioner will be able to take legal action against social media companies that have not pursued reasonable steps to bar users under the age of 16, and that fines can range up to A$49.5 million against these companies in court. On 19 November 2025, Meta announced that from 4 December their platforms (Instagram, Facebook, and Threads) would be removing users under the age of 16 ahead of the 10 December deadline. Users will be able to scan a face or provide an identity document to prove their age. On 21 November 2025, the eSafety Commissioner announced that the live-streaming platform Twitch will be included in the ban, but that Pinterest would not be. In December 2025, eSafety Commissioner Julie Inman Grant suggested efforts to block users include use by social media companies of various "signals" to identify children that are