In machine learning, a neural scaling law is an empirical scaling law that describes how neural network performance changes as key factors are scaled up or down. These factors typically include the number of parameters, training dataset size, and training cost. Some models also exhibit performance gains by scaling inference through increased test-time compute (TTC), extending neural scaling laws beyond training to the deployment phase. == Introduction == In general, a deep learning model can be characterized by four parameters: model size, training dataset size, training cost, and the post-training error rate (e.g., the test set error rate). Each of these variables can be defined as a real number, usually written as N , D , C , L {\displaystyle N,D,C,L} (respectively: parameter count, dataset size, computing cost, and loss). A neural scaling law is a theoretical or empirical statistical law between these parameters. There are also other parameters with other scaling laws. === Size of the model === In most cases, the model's size is simply the number of parameters. However, one complication arises with the use of sparse models, such as mixture-of-expert models. With sparse models, during inference, only a fraction of their parameters are used. In comparison, most other kinds of neural networks, such as transformer models, always use all their parameters during inference. === Size of the training dataset === The size of the training dataset is usually quantified by the number of data points within it. Larger training datasets are typically preferred, as they provide a richer and more diverse source of information from which the model can learn. This can lead to improved generalization performance when the model is applied to new, unseen data. However, increasing the size of the training dataset also increases the computational resources and time required for model training. With the "pretrain, then finetune" method used for most large language models, there are two kinds of training dataset: the pretraining dataset and the finetuning dataset. Their sizes have different effects on model performance. Generally, the finetuning dataset is less than 1% the size of pretraining dataset. In some cases, a small amount of high quality data suffices for finetuning, and more data does not necessarily improve performance. Many scaling laws, due to their inherent diminishing returns nature, value data based on a submodular set function which was shown in a paper on this topic. === Cost of training === Training cost is typically measured in terms of time (how long it takes to train the model) and computational resources (how much processing power and memory are required). It is important to note that the cost of training can be significantly reduced with efficient training algorithms, optimized software libraries, and parallel computing on specialized hardware such as GPUs or TPUs. The cost of training a neural network model is a function of several factors, including model size, training dataset size, the training algorithm complexity, and the computational resources available. In particular, doubling the training dataset size does not necessarily double the cost of training, because one may train the model for several times over the same dataset (each being an "epoch"). === Performance === The performance of a neural network model is evaluated based on its ability to accurately predict the output given some input data. Common metrics for evaluating model performance include: Negative log-likelihood per token (logarithm of perplexity) for language modeling; Accuracy, precision, recall, and F1 score for classification tasks; Mean squared error (MSE) or mean absolute error (MAE) for regression tasks; Elo rating in a competition against other models, such as gameplay or preference by a human judge. Performance can be improved by using more data, larger models, different training algorithms, regularizing the model to prevent overfitting, and early stopping using a validation set. When the performance is a number bounded within the range of [ 0 , 1 ] {\displaystyle [0,1]} , such as accuracy, precision, etc., it often scales as a sigmoid function of cost, as seen in the figures. == Examples == === (Hestness, Narang, et al, 2017) === The 2017 paper is a common reference point for neural scaling laws fitted by statistical analysis on experimental data. Previous works before the 2000s, as cited in the paper, were either theoretical or orders of magnitude smaller in scale. Whereas previous works generally found the scaling exponent to scale like L ∝ D − α {\displaystyle L\propto D^{-\alpha }} , with α ∈ { 0.5 , 1 , 2 } {\displaystyle \alpha \in \{0.5,1,2\}} , the paper found that α ∈ [ 0.07 , 0.35 ] {\displaystyle \alpha \in [0.07,0.35]} . Of the factors they varied, only task can change the exponent α {\displaystyle \alpha } . Changing the architecture optimizers, regularizers, and loss functions, would only change the proportionality factor, not the exponent. For example, for the same task, one architecture might have L = 1000 D − 0.3 {\displaystyle L=1000D^{-0.3}} while another might have L = 500 D − 0.3 {\displaystyle L=500D^{-0.3}} . They also found that for a given architecture, the number of parameters necessary to reach lowest levels of loss, given a fixed dataset size, grows like N ∝ D β {\displaystyle N\propto D^{\beta }} for another exponent β {\displaystyle \beta } . They studied machine translation with LSTM ( α ∼ 0.13 {\displaystyle \alpha \sim 0.13} ), generative language modelling with LSTM ( α ∈ [ 0.06 , 0.09 ] , β ≈ 0.7 {\displaystyle \alpha \in [0.06,0.09],\beta \approx 0.7} ), ImageNet classification with ResNet ( α ∈ [ 0.3 , 0.5 ] , β ≈ 0.6 {\displaystyle \alpha \in [0.3,0.5],\beta \approx 0.6} ), and speech recognition with two hybrid (LSTMs complemented by either CNNs or an attention decoder) architectures ( α ≈ 0.3 {\displaystyle \alpha \approx 0.3} ). === (Henighan, Kaplan, et al, 2020) === A 2020 analysis studied statistical relations between C , N , D , L {\displaystyle C,N,D,L} over a wide range of values and found similar scaling laws, over the range of N ∈ [ 10 3 , 10 9 ] {\displaystyle N\in [10^{3},10^{9}]} , C ∈ [ 10 12 , 10 21 ] {\displaystyle C\in [10^{12},10^{21}]} , and over multiple modalities (text, video, image, text to image, etc.). In particular, the scaling laws it found are (Table 1 of ): For each modality, they fixed one of the two C , N {\displaystyle C,N} , and varying the other one ( D {\displaystyle D} is varied along using D = C / 6 N {\displaystyle D=C/6N} ), the achievable test loss satisfies L = L 0 + ( x 0 x ) α {\displaystyle L=L_{0}+\left({\frac {x_{0}}{x}}\right)^{\alpha }} where x {\displaystyle x} is the varied variable, and L 0 , x 0 , α {\displaystyle L_{0},x_{0},\alpha } are parameters to be found by statistical fitting. The parameter α {\displaystyle \alpha } is the most important one. When N {\displaystyle N} is the varied variable, α {\displaystyle \alpha } ranges from 0.037 {\displaystyle 0.037} to 0.24 {\displaystyle 0.24} depending on the model modality. This corresponds to the α = 0.34 {\displaystyle \alpha =0.34} from the Chinchilla scaling paper. When C {\displaystyle C} is the varied variable, α {\displaystyle \alpha } ranges from 0.048 {\displaystyle 0.048} to 0.19 {\displaystyle 0.19} depending on the model modality. This corresponds to the β = 0.28 {\displaystyle \beta =0.28} from the Chinchilla scaling paper. Given fixed computing budget, optimal model parameter count is consistently around N o p t ( C ) = ( C 5 × 10 − 12 petaFLOP-day ) 0.7 = 9.0 × 10 − 7 C 0.7 {\displaystyle N_{opt}(C)=\left({\frac {C}{5\times 10^{-12}{\text{petaFLOP-day}}}}\right)^{0.7}=9.0\times 10^{-7}C^{0.7}} The parameter 9.0 × 10 − 7 {\displaystyle 9.0\times 10^{-7}} varies by a factor of up to 10 for different modalities. The exponent parameter 0.7 {\displaystyle 0.7} varies from 0.64 {\displaystyle 0.64} to 0.75 {\displaystyle 0.75} for different modalities. This exponent corresponds to the ≈ 0.5 {\displaystyle \approx 0.5} from the Chinchilla scaling paper. It's "strongly suggested" (but not statistically checked) that D o p t ( C ) ∝ N o p t ( C ) 0.4 ∝ C 0.28 {\displaystyle D_{opt}(C)\propto N_{opt}(C)^{0.4}\propto C^{0.28}} . This exponent corresponds to the ≈ 0.5 {\displaystyle \approx 0.5} from the Chinchilla scaling paper. The scaling law of L = L 0 + ( C 0 / C ) 0.048 {\displaystyle L=L_{0}+(C_{0}/C)^{0.048}} was confirmed during the training of GPT-3 (Figure 3.1 ). === Chinchilla scaling (Hoffmann, et al, 2022) === One particular scaling law ("Chinchilla scaling") states that, for a large language model (LLM) autoregressively trained for one epoch, with a cosine learning rate schedule, we have: { C = C 0 N D L = A N α + B D β + L 0 {\displaystyle {\begin{cases}C=C_{0}ND\\L={\frac {A}{N^{\alpha }}}+{\frac {B}{D^{\beta }}}+L_{0}\end{cases}}} where the variables are C {\displaystyle C} is the cost o
Afghan Girls Robotics Team
The Afghan Girls Robotics Team, also known as the Afghan Dreamers, is an all-girl robotics team from Herat, Afghanistan, founded through the Digital Citizen Fund (DCF) in 2017 by Roya Mahboob and Alireza Mehraban. It is made up of girls between ages 12 and 18 and their mentors. Several members of the team were relocated to Qatar and Mexico by humanitarian and tech entrepreneur Sarah Porter following the fall of Kabul in August 2021. A documentary film featuring members of the team, titled Afghan Dreamers, was released by MTV Documentary Films in 2023. == Origins == The Afghan Girls Robotics Team was co-founded in 2017 by Roya Mahboob, who is their coach, mentor and sponsor, and founder of the Digital Citizen Fund (DCF), which is the parent organization for the team. Dean Kamen was planning a 2017 competition in the United States and had recruited Mahboob to form a team from Afghanistan. Out of 150 girls, 12 were selected for the first team. Before parts were sent by Kamen, they trained in the basement of the home of Mahboob's parents, with scrap metal and without safety equipment under the guidance of their coach, Mahboob's brother Alireza Mehraban, who is also a co-founder of the team. == 2017 and 2018 == In 2017, six members of the Afghan Girls Robotics Team traveled to the United States to participate in the international FIRST Global Challenge robotics competition. Their visas were rejected twice after they made two journeys from Herat to Kabul through Taliban-controlled areas, before officials in the United States government intervened to allow them to enter the United States. Customs officials also detained their robotics kits, which left them two weeks to construct their robot, unlike some teams that had more time. They were awarded a Silver medal for Courageous Achievement. One week after they returned home from the competition, the father of team captain Fatemah Qaderyan, Mohammad Asif Qaderyan, was killed in a suicide bombing. After their United States visas expired, the team participated in competitions in Estonia and Istanbul. Three of the 12 members participated in the 2017 Entrepreneurial Challenge at the Robotex festival in Estonia, and won the competition for their solar-powered robot designed to assist farmers. In 2018, the team trained in Canada, continued to travel in the United States for months and participate in competitions. == 2019 == The Afghan Girls Robotics team had aspirations to develop a science and technology school for girls in Afghanistan. Roya Mahboob interfaced with the School of Engineering and Applied Sciences (SEAS), the School of Architecture, and the Whitney and Betty MacMillan Center for International and Area Studies Yale University to design the infrastructure for what they named The Dreamer Institute. == 2020 == In March 2020, the governor of Herat at the time, in response to the COVID-19 pandemic in Afghanistan and a scarcity of ventilators, sought help with the design of low-cost ventilators, and the Afghan Girls Robotics Team was one of six teams contacted by the government. Using a design from Massachusetts Institute of Technology and with guidance from MIT engineers and Douglas Chin, a surgeon in California, the team developed a prototype with Toyota Corolla parts and a chain drive from a Honda motorcycle. UNICEF also supported the team with the acquisition of necessary parts during the three months they spent building the prototype that was completed in July 2020. Their design costs around $500 compared to $50,000 for a ventilator. In December 2020, Minister of Industry and Commerce Nizar Ahmad Ghoryani donated funding and obtained land for a factory to produce the ventilators. Under the direction of their mentor Roya Mahboob, the Afghan Dreamers also designed a UVC Robot for sanitization, and a Spray Robot for disinfection, both of which were approved by the Ministry of Health for production. == 2021 == In early August 2021, Somaya Faruqi, former captain of the team, was quoted by Public Radio International about the future of Afghanistan, stating, "We don’t support any group over another but for us what’s important is that we be able to continue our work. Women in Afghanistan have made a lot of progress over the past two decades and this progress must be respected." On August 17, 2021, the Afghan Girls Robotics Team and their coaches were reported to be attempting to evacuate, but unable to obtain a flight out of Afghanistan, and a lawyer appealed to Canada for assistance regarding the evacuation of the team members. As of August 19, 2021, nine members of the team and their coaches had evacuated to Qatar. The founder of the team, Roya Mahboob, and DCF board member, Elizabeth Schaeffer Brown, were previously in contact with the Qatari government to assist the team members in their evacuation from Afghanistan. By August 25, 2021, some members arrived in Mexico. Saghar, a team member who evacuated to Mexico, said, "We wanted to continue the path that we started to continue to go for our achievements and to go for having our dreams through reality. So that's why we decided to leave Afghanistan and go for somewhere safe" in an interview with The Associated Press. The members who have left Afghanistan participated in an online robotics competition in September and plan to continue their education. A documentary film titled Afghan Dreamers, produced by Beth Murphy and directed by David Greenwald, was in post-production when the team began to evacuate. == 2022 == The Afghan Dreamers were involved in a training program at the Texas A&M University at Qatar’s STEM Hub. == 2023 == The Afghan Girls Robotics Team had a booth at the 5th UN Conference on the Least Developed Countries, where they displayed some of the robots the team had constructed. == Afghan Dreamers documentary == The Afghan Dreamers documentary from MTV Documentary Films premiered in May 2023 on Paramount+. The film was directed by David Greenwald and produced by David Cowan and Beth Murphy. In a review for Screen Daily, Wendy Ide wrote, "This film, with its likeable cast of girl nerds and positive message, should enjoy a warm reception on the festival circuit, and will be of particular interest to events seeking to showcase women's stories from around the world. It also serves as a timely cautionary tale – a case study on just how quickly the rights and the opportunities of women can be curtailed, at the behest of the men in power." == Honors and awards == 2017 Silver medal for Courageous Achievement at the FIRST Global Challenge, science and technology 2017 Benefiting Humanity in AI Award at World Summit AI 2017 Winner, Entrepreneurship Challenge at Robotex in Estonia 2018 Permission to Dream Award, Raw Film Festival 2018 Conrad Innovation Challenge, Raw Film Festival 2018 Rookie All Star – District Championship, Canada 2018 Asia Game Changer Award Honoree 2019 Inspiring in Engineering Award – FIRST Detroit World Championship 2019 Asia Game Changer Award of California 2019 Safety Award – FIRST Global, Dubai 2021 Forbes 30 Under 30 Asia 2022 World Championships, Genoa, Switzerland
Law practice management software
Law practice management software is software designed to manage the business operations of a law firm. This can include software that manages cases, client intake, court communications, electronic discovery, time tracking, trust accounting, and billing. == Features of law practice management software == Common features of practice management software include: Case management Time tracking Document assembly Contact management Calendaring Docket management Client portal Contract Management Court Case Status Tracker Trust accounting == Examples of law practice management software == Smokeball LEAP Legal Software PracticeEvolve Dye & Durham
Ogle app
Ogle is a free smartphone based social media application. It is available for iOS and Android. Ogle acts like a school wide forum that lets users and users' classmates share and interact. Users can share photos, videos, questions, even thoughts and watch submissions grow in popularity as other users vote and comment on them. == App Features == Campus Feed: Interact by watching and posting videos or pictures to your campus story. Photos and Videos: share what you want with many different timing options. Interact: Chat with friends and groups, or share a moment for all to see. Real-name system: choose to register an account with username and profile picture. Custom Stickers: Create stickers to add creativity and zest to your pictures. Flash Interaction: All private chat and group chat history will be deleted after 24 hours on Ogle Chat. == Controversies == Users can post anything on Ogle using text, photos, and videos. As a result, some Ogle user's sense of anonymity, posts have targeted specific schools and students with abusive and hurtful content. The Ogle app's user anonymity makes it difficult for school officials to quickly investigate issues that occur within the Ogle app. On March 28, 2016, three people were arrested after violent threats were made against an Anaheim high school. 18-year-old Miguel Meza was arrested Sunday afternoon during a traffic stop, along with his passenger, 23-year-old Johnny Aguilar. Police said both men had loaded handguns. Aguilar was also accused of violating his probation. "It is concerning the fact that they did have firearms, but we don't have a crystal ball. We can't determine if they possessed those firearms to engage in some kind of school violence or if they had it for another reason," Sgt. Daron Wyatt with the Anaheim Police Department said. Officials said Meza and Aguilar have known gang ties and detectives began investigating Meza after threats were made against the school on Ogle. On February 29, 2016, Santa Cruz County sheriff's deputies arrested a 16-year-old Aptos High School student Friday, accused of making an online threat of gun violence at Aptos High and Monte Vista Christian."He basically told detectives that it was all a joke. It's not a joke. You have multiple resources being spent to investigate these cases," said Santa Cruz County Sheriff's Sgt. Roy Morales. The schools remained open throughout the week, with a huge police presence on campus. In an anonymous emailed statement to the Daily Pilot on Thursday, the "Ogle team" said: "We are aware of the concern, and cyberbullying is absolutely NOT our intention for the app. Our goal for this app is to create a free and safe community space for students, for a better communication. We are currently working around the clock to improve the app. As a matter of fact, we are also in contact with local police departments, anti-bullying organizations and local high schools to try to help the students." In response to these incidents, Ogle expressed that they takes the safety of its users seriously and does not condone any type of behavior that is illegal or in violation of its content policies. The company also said it has instituted a content moderation team to increase review and identify and remove inappropriate content, and take action against “those who violate our community guidelines.”
Robot Monk Xian'er
Robot Monk Xian'er (Chinese: 贤二机器僧) is a humanoid robot based on the cartoon character Xian'er. It was developed by a team of monks, volunteers and AI experts from Beijing Longquan Monastery in Beijing, China. He can follow human instructions to make body movements, read scriptures and play Buddhist music. He can chat and respond to people's emotional and spiritual questions with Buddhist wisdom. As a chatbot, Robot Monk Xian'er is available on certain public platforms including WeChat and Facebook. Over the years, master Xuecheng, the abbot of Beijing Longquan Monastery, replied to thousands of questions on Sina Weibo. These questions and their answers become the data source of the chatbot.
Seq2seq
Seq2seq is a family of machine learning approaches used for natural language processing. Originally developed by Lê Viết Quốc, a Vietnamese computer scientist and a machine learning pioneer at Google Brain, this framework has become foundational in many modern AI systems. Applications include language translation, image captioning, conversational models, speech recognition, and text summarization. Seq2seq uses sequence transformation: it turns one sequence into another sequence. == History == One naturally wonders if the problem of translation could conceivably be treated as a problem in cryptography. When I look at an article in Russian, I say: 'This is really written in English, but it has been coded in some strange symbols. I will now proceed to decode. seq2seq is an approach to machine translation (or more generally, sequence transduction) with roots in information theory, where communication is understood as an encode-transmit-decode process, and machine translation can be studied as a special case of communication. This viewpoint was elaborated, for example, in the noisy channel model of machine translation. In practice, seq2seq maps an input sequence into a real-numerical vector by using a neural network (the encoder), and then maps it back to an output sequence using another neural network (the decoder). The idea of encoder-decoder sequence transduction had been developed in the early 2010s. The papers most commonly cited as the originators that produced seq2seq are two papers from 2014. In the seq2seq as proposed by them, both the encoder and the decoder were LSTMs. This had the "bottleneck" problem, since the encoding vector has a fixed size, so for long input sequences, information would tend to be lost, as they are difficult to fit into the fixed-length encoding vector. The attention mechanism, proposed in 2014, resolved the bottleneck problem. They called their model RNNsearch, as it "emulates searching through a source sentence during decoding a translation". A problem with seq2seq models at this point was that recurrent neural networks are difficult to parallelize. The 2017 publication of Transformers resolved the problem by replacing the encoding RNN with self-attention Transformer blocks ("encoder blocks"), and the decoding RNN with cross-attention causally-masked Transformer blocks ("decoder blocks"). === Priority dispute === One of the papers cited as the originator for seq2seq is (Sutskever et al 2014), published at Google Brain while they were on Google's machine translation project. The research allowed Google to overhaul Google Translate into Google Neural Machine Translation in 2016. Tomáš Mikolov claims to have developed the idea (before joining Google Brain) of using a "neural language model on pairs of sentences... and then [generating] translation after seeing the first sentence"—which he equates with seq2seq machine translation, and to have mentioned the idea to Ilya Sutskever and Quoc Le (while at Google Brain), who failed to acknowledge him in their paper. Mikolov had worked on RNNLM (using RNN for language modelling) for his PhD thesis, and is more notable for developing word2vec. == Architecture == The main reference for this section is. === Encoder === The encoder is responsible for processing the input sequence and capturing its essential information, which is stored as the hidden state of the network and, in a model with attention mechanism, a context vector. The context vector is the weighted sum of the input hidden states and is generated for every time instance in the output sequences. === Decoder === The decoder takes the context vector and hidden states from the encoder and generates the final output sequence. The decoder operates in an autoregressive manner, producing one element of the output sequence at a time. At each step, it considers the previously generated elements, the context vector, and the input sequence information to make predictions for the next element in the output sequence. Specifically, in a model with attention mechanism, the context vector and the hidden state are concatenated together to form an attention hidden vector, which is used as an input for the decoder. The seq2seq method developed in the early 2010s uses two neural networks: an encoder network converts an input sentence into numerical vectors, and a decoder network converts those vectors to sentences in the target language. The Attention mechanism was grafted onto this structure in 2014 and is shown below. Later it was refined into the encoder-decoder Transformer architecture of 2017. === Training vs prediction === There is a subtle difference between training and prediction. During training time, both the input and the output sequences are known. During prediction time, only the input sequence is known, and the output sequence must be decoded by the network itself. Specifically, consider an input sequence x 1 : n {\displaystyle x_{1:n}} and output sequence y 1 : m {\displaystyle y_{1:m}} . The encoder would process the input x 1 : n {\displaystyle x_{1:n}} step by step. After that, the decoder would take the output from the encoder, as well as the
Common data model
A common data model (CDM) can refer to any standardised data model which allows for data and information exchange between different applications and data sources. Common data models aim to standardise logical infrastructure so that related applications can "operate on and share the same data", and can be seen as a way to "organize data from many sources that are in different formats into a standard structure". A common data model has been described as one of the components of a "strong information system". A standardised common data model has also been described as a typical component of a well designed agile application besides a common communication protocol. Providing a single common data model within an organisation is one of the typical tasks of a data warehouse. == Examples of common data models == === Border crossings === X-trans.eu was a cross-border pilot project between the Free State of Bavaria (Germany) and Upper Austria with the aim of developing a faster procedure for the application and approval of cross-border large-capacity transports. The portal was based on a common data model that contained all the information required for approval. === Climate data === The Climate Data Store Common Data Model is a common data model set up by the Copernicus Climate Change Service for harmonising essential climate variables from different sources and data providers. === General information technology === Within service-oriented architecture, S-RAMP is a specification released by HP, IBM, Software AG, TIBCO, and Red Hat which defines a common data model for SOA repositories as well as an interaction protocol to facilitate the use of common tooling and sharing of data. Content Management Interoperability Services (CMIS) is an open standard for inter-operation of different content management systems over the internet, and provides a common data model for typed files and folders used with version control. The NetCDF software libraries for array-oriented scientific data implements a common data model called the NetCDF Java common data model, which consists of three layers built on top of each other to add successively richer semantics. === Health === Within genomic and medical data, the Observational Medical Outcomes Partnership (OMOP) research program established under the U.S. National Institutes of Health has created a common data model for claims and electronic health records which can accommodate data from different sources around the world. PCORnet, which was developed by the Patient-Centered Outcomes Research Institute, is another common data model for health data including electronic health records and patient claims. The Sentinel Common Data Model was initially started as Mini-Sentinel in 2008. It is used by the Sentinel Initiative of the USA's Food and Drug Administration. The Generalized Data Model was first published in 2019. It was designed to be a stand-alone data model as well as to allow for further transformation into other data models (e.g., OMOP, PCORNet, Sentinel). It has a hierarchical structure to flexibly capture relationships among data elements. The JANUS clinical trial data repository also provides a common data model which is based on the SDTM standard to represent clinical data submitted to regulatory agencies, such as tabulation datasets, patient profiles, listings, etc. === Logistics === SX000i is a specification developed jointly by the Aerospace and Defence Industries Association of Europe (ASD) and the American Aerospace Industries Association (AIA) to provide information, guidance and instructions to ensure compatibility and the commonality. The associated SX002D specification contains a common data model. === Microsoft Common Data Model === The Microsoft Common Data Model is a collection of many standardised extensible data schemas with entities, attributes, semantic metadata, and relationships, which represent commonly used concepts and activities in various businesses areas. It is maintained by Microsoft and its partners, and is published on GitHub. Microsoft's Common Data Model is used amongst others in Microsoft Dataverse and with various Microsoft Power Platform and Microsoft Dynamics 365 services. === Rail transport === RailTopoModel is a common data model for the railway sector. === Other === There are many more examples of various common data models for different uses published by different sources.