Model compression is a machine learning technique for reducing the size of trained models. Large models can achieve high accuracy, but often at the cost of significant resource requirements. Compression techniques aim to compress models without significant performance reduction. Smaller models require less storage space, and consume less memory and compute during inference. Compressed models enable deployment on resource-constrained devices such as smartphones, embedded systems, edge computing devices, and consumer electronics computers. Efficient inference is also valuable for large corporations that serve large model inference over an API, allowing them to reduce computational costs and improve response times for users. Model compression is not to be confused with knowledge distillation, in which a smaller "student" model is trained to imitate the input-output behavior of a larger "teacher" model (as opposed to using the "teacher"'s trained parameters or the "teacher"'s training targets). == Techniques == Several techniques are employed for model compression. === Pruning === Pruning sparsifies a large model by setting some parameters to exactly zero. This effectively reduces the number of parameters. This allows the use of sparse matrix operations, which are faster than dense matrix operations. Pruning criteria can be based on magnitudes of parameters, the statistical pattern of neural activations, Hessian values, etc. === Quantization === Quantization reduces the numerical precision of weights and activations. For example, instead of storing weights as 32-bit floating-point numbers, they can be represented using 8-bit integers. Low-precision parameters take up less space, and takes less compute to perform arithmetic with. It is also possible to quantize some parameters more aggressively than others, so for example, a less important parameter can have 8-bit precision while another, more important parameter, can have 16-bit precision. Inference with such models requires mixed-precision arithmetic. Quantized models can also be used during training (rather than after training). PyTorch implements automatic mixed-precision (AMP), which performs autocasting, gradient scaling, and loss scaling. === Low-rank factorization === Weight matrices can be approximated by low-rank matrices. Let W {\displaystyle W} be a weight matrix of shape m × n {\displaystyle m\times n} . A low-rank approximation is W ≈ U V T {\displaystyle W\approx UV^{T}} , where U {\displaystyle U} and V {\displaystyle V} are matrices of shapes m × k , n × k {\displaystyle m\times k,n\times k} . When k {\displaystyle k} is small, this both reduces the number of parameters needed to represent W {\displaystyle W} approximately, and accelerates matrix multiplication by W {\displaystyle W} . Low-rank approximations can be found by singular value decomposition (SVD). The choice of rank for each weight matrix is a hyperparameter, and jointly optimized as a mixed discrete-continuous optimization problem. The rank of weight matrices may also be pruned after training, taking into account the effect of activation functions like ReLU on the implicit rank of the weight matrices. == Training == Model compression may be decoupled from training, that is, a model is first trained without regard for how it might be compressed, then it is compressed. However, it may also be combined with training. The "train big, then compress" method trains a large model for a small number of training steps (less than it would be if it were trained to convergence), then heavily compress the model. It is found that at the same compute budget, this method results in a better model than lightly compressed, small models. In Deep Compression, the compression has three steps. First loop (pruning): prune all weights lower than a threshold, then finetune the network, then prune again, etc. Second loop (quantization): cluster weights, then enforce weight sharing among all weights in each cluster, then finetune the network, then cluster again, etc. Third step: Use Huffman coding to losslessly compress the model. The SqueezeNet paper reported that Deep Compression achieved a compression ratio of 35 on AlexNet, and a ratio of ~10 on SqueezeNets.
TigerGraph
TigerGraph is a private company headquartered in Redwood City, California. It provides graph database and graph analytics software. == History == TigerGraph was founded in 2012 by programmer Yu, Ruoming, Li, Like and Mingxi, under the name GraphSQL. In September 2017, the company came out of stealth mode under the name TigerGraph with $33 million in funding. It raised an additional $32 million in funding in September 2019 and another $105 million in a series C round in February 2021. Cumulative funding as of March 2021 is $170 million. == Products == TigerGraph's hybrid transactional/analytical processing database and analytics software can scale to hundreds of terabytes of data with trillions of edges, and is used for data intensive applications such as fraud detection, customer data analysis (customer 360), IoT, artificial intelligence and machine learning. It is available using the cloud computing delivery model. The analytics uses C++ based software and a parallel processing engine to process algorithms and queries. It has its own graph query language that is similar to SQL. TigerGraph also provides a software development kit for creating graphs and visual representations. As of Mar 2024, TigerGraph version is up to version 4.2.0 TigerGraph offers free Community Edition for developers, researchers, and educators. It can be obtained from https://dl.tigergraph.com/ == Query Language == GSQL , designed by Mingxi Wu and Alin Deutsch in 2015, is a SQL-like Turing complete query language. GSQL includes additions to make it compliant with the Graph Query Language standard.
Hancom Office
Hancom Office is a proprietary office suite that includes a word processor, spreadsheet software, presentation software, and a PDF editor as well as their online versions accessible via a web browser. It is primarily addressed to Korean users. Hancom Office is written in Java and C++ that runs on Android, iOS, macOS and Windows platforms. == Products == Hangul - Hangul is a word processor developed by Hancom. It is a product that eliminates the inconvenience of the original Hangul word processor, which was limited to Hangul cards or PC models. Originally, the name was written using the '아래아' character, a vowel letter that is obsolete in modern Korean, and it was referred to as 'HWP' (an abbreviation for Hangul Word Processor), '아래아 한글' (Arae-a Hangul), '한/글' (Han/Geul), and so on. Hangul is currently the most widely used word processor in South Korea, often used alongside Microsoft Word. HanWord - word processor compatible with Word HanCell - spreadsheet program HanShow - presentation program Hancom Office Hanword Viewer - For viewing documents created by Hancom Office or Microsoft Office
Software configuration management
Software configuration management (SCM), a.k.a. software change and configuration management (SCCM), is the software engineering practice of tracking and controlling changes to a software system. It is part of the larger cross-disciplinary field of configuration management (CM). SCM includes version control and the establishment of baselines. == Goals == The goals of SCM include: Configuration identification - Identifying configurations, configuration items and baselines. Configuration control - Implementing a controlled change process. This is usually achieved by setting up a change control board whose primary function is to approve or reject all change requests that are sent against any baseline. Configuration status accounting - Recording and reporting all the necessary information on the status of the development process. Configuration auditing - Ensuring that configurations contain all their intended parts and are sound with respect to their specifying documents, including requirements, architectural specifications and user manuals. Build management - Managing the process and tools used for builds. Process management - Ensuring adherence to the organization's development process. Environment management - Managing the software and hardware that host the system. Teamwork - Facilitate team interactions related to the process. Defect tracking - Making sure every defect has traceability back to the source. With the introduction of cloud computing and DevOps the purposes of SCM tools have become merged in some cases. The SCM tools themselves have become virtual appliances that can be instantiated as virtual machines and saved with state and version. The tools can model and manage cloud-based virtual resources, including virtual appliances, storage units, and software bundles. The roles and responsibilities of the actors have become merged as well with developers now being able to dynamically instantiate virtual servers and related resources. == History == == Examples == Ansible – Open-source software platform for remote configuring and managing computers CFEngine – Configuration management software Chef – Configuration management toolPages displaying short descriptions of redirect targets LCFG – Computer configuration management system NixOS – Linux distribution OpenMake Software – DevOps company Otter Puppet – Open source configuration management software Salt – Configuration management software Rex – Open source software
How Data Happened
How Data Happened: A History from the Age of Reason to the Age of Algorithms is a 2023 non-fiction book written by Columbia University professors Chris Wiggins and Matthew L. Jones. The book explores the history of data and statistics from the end of the 18th century to the present day. == Content == The book starts at the end of the 18th century, when European states began tabulating physical resources, and ends at the present day, when algorithms manipulate our personal information as a commodity. It looks at the rise of data and statistics, and how early statistical methods were used to justify eugenics, quantify supposed racial differences, and develop military and industrial applications. The authors also discuss the impact of the internet and e-commerce on data collection, the rise of data science, and the consequences of government-run surveillance systems collecting vast amounts of personal data for customized, targeted advertising. They emphasize the importance of privacy and democracy and propose remedies to the problems caused by mass data collection, including stronger regulation of the tech industry and collective action by its employees. The book is a historical analysis that provides context for understanding the debates surrounding data and its control. The book has 336 pages and was published in 2023 by W. W. Norton & Company.
Tapingo
Tapingo was an American mobile commerce application that offers advance ordering for pickup and food delivery services for college campuses. The company was acquired by Grubhub in September 2018 for approximately $150 million. Following the acquisition, Tapingo’s campus-ordering functionality was integrated into the Grubhub app (Grubhub Campus Dining) and the Tapingo service was discontinued during 2019. Tapingo is differentiated from other on-demand delivery/logistics companies, such as Waiter.com, Postmates, or DoorDash, by focusing its efforts on serving the college market. Through Tapingo, users can browse menus, place orders, pay for the meal and schedule the pickup or have it delivered. On certain campuses, students are able to use their university's meal dollars to pay for food. In the spring of 2012, Tapingo first launched its services on five campuses (Santa Clara University, Loyola Marymount University, Biola University, the University of Maine, and California Lutheran University), and has since expanded to more than 200 college campuses across the U.S. and Canada, serving 100 markets. To date, Tapingo has received venture funding from Carmel Ventures, Khosla Ventures, Kinzon Capital, DCM Ventures and Qualcomm Ventures. In fall 2015, Tapingo announced expansion plans through major partnership deals with national brands like Chipotle Mexican Grill and 7-Eleven, regional restaurants such as Taco Bueno, and global foodservice provider Aramark.
Google Cloud Dataflow
Google Cloud Dataflow is a fully managed service for executing Apache Beam pipelines within the Google Cloud Platform ecosystem. Dataflow provides a fully managed service for executing Apache Beam pipelines, offering features like autoscaling, dynamic work rebalancing, and a managed execution environment. Dataflow is suitable for large-scale, continuous data processing jobs, and is one of the major components of Google's big data architecture on the Google Cloud Platform. At its core, Dataflow's architecture is designed to abstract away infrastructure management, allowing developers to focus purely on the logic of their data processing tasks. When a pipeline written using the Apache Beam SDK is submitted, Dataflow translates this high-level definition into an optimized job graph. The service then provisions and manages a fleet of Google Compute Engine workers to execute this graph in a highly parallelized and fault-tolerant manner. This serverless approach, combined with intelligent autoscaling of both the number of workers (horizontal) and the resources per worker (vertical), ensures that jobs have the precise amount of computational power needed at any given time, optimizing both performance and cost. The service's deep integration with the Google Cloud ecosystem makes it a powerful tool for a variety of use cases beyond simple data movement. For real-time analytics, Dataflow can ingest unbounded streams of data from Cloud Pub/Sub, perform complex transformations, and load results into BigQuery for immediate querying. In machine learning workflows, it is commonly used to preprocess and transform massive datasets stored in Cloud Storage, preparing them for training models in Vertex AI. This versatility makes it the central processing engine for modern ETL (Extract, Transform, Load) operations, streaming analytics, and large-scale data preparation within the cloud. == History == Google Cloud Dataflow was announced in June, 2014 and released to the general public as an open beta in April, 2015. In January, 2016 Google donated the underlying SDK, the implementation of a local runner, and a set of IOs (data connectors) to access Google Cloud Platform data services to the Apache Software Foundation. The donated code formed the original basis for Apache Beam. In August 2022, there was an incident where user timers were broken for certain Dataflow streaming pipelines in multiple regions, which was later resolved. Throughout 2023 and 2024, there have been various other updates and incidents affecting Google Cloud Dataflow, as documented in the release notes and service health history. The donation of the Dataflow SDK to the Apache Software Foundation was a pivotal moment, establishing Apache Beam as a unified, open-source programming model for defining both batch and streaming data pipelines. This strategic move decoupled the pipeline definition from the execution engine. As a result, developers could write portable data processing logic that was not locked into Google's ecosystem. A Beam pipeline can be executed on various runners, including Apache Flink, Apache Spark, and, of course, the highly optimized Google Cloud Dataflow service, providing flexibility and future-proofing data processing investments. == Features == Google Cloud Dataflow supports both batch and streaming data processing pipelines. It automatically handles resource provisioning, data sharding, and scaling according to workload, reducing manual configuration needed for large-scale data operations. == Use cases == Dataflow is used for ETL (Extract, Transform, Load) data pipelines, real-time analytics, and event stream processing for companies in industries such as finance, advertising, and IoT.