Merit Network, Inc., is a nonprofit member-governed organization providing high-performance computer networking and related services to educational, government, health care, and nonprofit organizations, primarily in Michigan. Created in 1966, Merit operates the longest running regional computer network in the United States. == Organization == Created in 1966 as the Michigan Educational Research Information Triad by Michigan State University (MSU), the University of Michigan (U-M), and Wayne State University (WSU), Merit was created to investigate resource sharing by connecting the mainframe computers at these three Michigan public research universities. Merit's initial three node packet-switched computer network was operational in October 1972 using custom hardware based on DEC PDP-11 minicomputers and software developed by the Merit staff and the staffs at the three universities. Over the next dozen years the initial network grew as new services such as dial-in terminal support, remote job submission, remote printing, and file transfer were added; as gateways to the national and international Tymnet, Telenet, and Datapac networks were established, as support for the X.25 and TCP/IP protocols was added; as additional computers such as WSU's MVS system and the UM's electrical engineering's VAX running UNIX were attached; and as new universities became Merit members. Merit's involvement in national networking activities started in the mid-1980s with connections to the national supercomputing centers and work on the 56 kbit/s National Science Foundation Network (NSFNET), the forerunner of today's Internet. From 1987 until April 1995, Merit re-engineered and managed the NSFNET backbone service. MichNet, Merit's regional network in Michigan was attached to NSFNET and in the early 1990s Merit began extending "the Internet" throughout Michigan, offering both direct connect and dial-in services, and upgrading the statewide network from 56 kbit/s to 1.5 Mbit/s, and on to 45, 155, 622 Mbit/s, and eventually 1 and 10 Gbit/s. In 2003 Merit began its transition to a facilities based network, using fiber optic facilities that it shares with its members, that it purchases or leases under long-term agreements, or that it builds. In addition to network connectivity services, Merit offers a number of related services within Michigan and beyond, including: Internet2 connectivity, VPN, Network monitoring, Voice over IP (VOIP), Cloud storage, E-mail, Domain Name, Network Time, VMware and Zimbra software licensing, Colocation, and professional development seminars, workshops, classes, conferences, and meetings. == History == === Creating the network: 1966 to 1973 === The Michigan Educational Research Information Triad (MERIT) was formed in the fall of 1966 by Michigan State University (MSU), University of Michigan (U-M), and Wayne State University (WSU). More often known as the Merit Computer Network or simply Merit, it was created to design and implement a computer network connecting the mainframe computers at the universities. In the fall of 1969, after funding for the initial development of the network had been secured, Bertram Herzog was named director for MERIT. Eric Aupperle was hired as senior engineer, and was charged with finding hardware to make the network operational. The National Science Foundation (NSF) and the State of Michigan provided the initial funding for the network. In June 1970, the Applied Dynamics Division of Reliance Electric in Saline, Michigan was contracted to build three Communication Computers or CCs. Each would consist of a Digital Equipment Corporation (DEC) PDP-11 computer, dataphone interfaces, and interfaces that would attach them directly to the mainframe computers. The cost was to be slightly less than the $300,000 ($2,487,100, adjusted for inflation) originally budgeted. Merit staff wrote the software that ran on the CCs, while staff at each of the universities wrote the mainframe software to interface to the CCs. The first completed connection linked the IBM S/360-67 mainframe computers running the Michigan Terminal System at WSU and U-M, and was publicly demonstrated on December 14, 1971. The MSU node was completed in October 1972, adding a CDC 6500 mainframe running Scope/Hustler. The network was officially dedicated on May 15, 1973. === Expanding the network: 1974 to 1985 === In 1974, Herzog returned to teaching in the University of Michigan's Industrial Engineering Department, and Aupperle was appointed as director. Use of the all uppercase name "MERIT" was abandoned in favor of the mixed case "Merit". The first network connections were host to host interactive connections which allowed person to remote computer or local computer to remote computer interactions. To this, terminal to host connections, batch connections (remote job submission, remote printing, batch file transfer), and interactive file copy were added. And, in addition to connecting to host computers over custom hardware interfaces, the ability to connect to hosts or other networks over groups of asynchronous ports and via X.25 were added. Merit interconnected with Telenet (later SprintNet) in 1976 to give Merit users dial-in access from locations around the United States. Dial-in access within the U.S. and internationally was further expanded via Merit's interconnections to Tymnet, ADP's Autonet, and later still the IBM Global Network as well as Merit's own expanding network of dial-in sites in Michigan, New York City, and Washington, D.C. In 1978, Western Michigan University (WMU) became the fourth member of Merit (prompting a name change, as the acronym Merit no longer made sense as the group was no longer a triad). To expand the network, the Merit staff developed new hardware interfaces for the Digital PDP-11 based on printed circuit technology. The new system became known as the Primary Communications Processor (PCP), with the earliest PCPs connecting a PDP-10 located at WMU and a DEC VAX running UNIX at U-M's Electrical Engineering department. A second hardware technology initiative in 1983 produced the smaller Secondary Communication Processors (SCP) based on DEC LSI-11 processors. The first SCP was installed at the Michigan Union in Ann Arbor, creating UMnet, which extended Merit's network connectivity deeply into the U-M campus. In 1983 Merit's PCP and SCP software was enhanced to support TCP/IP and Merit interconnected with the ARPANET. === National networking, NSFNET, and the Internet: 1986 to 1995 === In 1986 Merit engineered and operated leased lines and satellite links that allowed the University of Michigan to access the supercomputing facilities at Pittsburgh, San Diego, and NCAR. In 1987, Merit, IBM and MCI submitted a winning proposal to NSF to implement a new NSFNET backbone network. The new NSFNET backbone network service began July 1, 1988. It interconnected supercomputing centers around the country at 1.5 megabits per second (T1), 24 times faster than the 56 kilobits-per-second speed of the previous network. The NSFNET backbone grew to link scientists and educators on university campuses nationwide and connect them to their counterparts around the world. The NSFNET project caused substantial growth at Merit, nearly tripling the staff and leading to the establishment of a new 24-hour Network Operations Center at the U-M Computer Center. In September 1990 in anticipation of the NSFNET T3 upgrade and the approaching end of the 5-year NSFNET cooperative agreement, Merit, IBM, and MCI formed Advanced Network and Services (ANS), a new non-profit corporation with a more broadly based Board of Directors than the Michigan-based Merit Network. Under its cooperative agreement with NSF, Merit remained ultimately responsible for the operation of NSFNET, but subcontracted much of the engineering and operations work to ANS. In 1991 the NSFNET backbone service was expanded to additional sites and upgraded to a more robust 45 Mbit/s (T3) based network. The new T3 backbone was named ANSNet and provided the physical infrastructure used by Merit to deliver the NSFNET Backbone Service. On April 30, 1995, the NSFNET project came to an end, when the NSFNET backbone service was decommissioned and replaced by a new Internet architecture with commercial Internet service providers (ISPs) interconnected at Network Access Points provided by multiple providers across the country. === Bringing the Internet to Michigan: 1985 to 2001 === During the 1980s, Merit Network grew to serve eight member universities, with Oakland University joining in 1985 and Central Michigan University, Eastern Michigan University, and Michigan Technological University joining in 1987. In 1990, Merit's board of directors formally changed the organization's name to Merit Network, Inc., and created the name MichNet to refer to Merit's statewide network. The board also approved a staff proposal to allow organizations other than publicly supported universities, referred to as aff
Association for Computational Linguistics
The Association for Computational Linguistics (ACL) is a scientific and professional organization for people working on natural language processing. Its namesake conference is one of the primary high impact conferences for natural language processing research, along with EMNLP. The conference is held each summer in locations where significant computational linguistics research is carried out. It was founded in 1962, originally named the Association for Machine Translation and Computational Linguistics (AMTCL). It became the ACL in 1968. The ACL has a European (EACL), a North American (NAACL), and an Asian (AACL) chapter. == History == The ACL was founded in 1962 as the Association for Machine Translation and Computational Linguistics (AMTCL). The initial membership was about 100. In 1965, the AMTCL took over the journal Mechanical Translation and Computational Linguistics. This journal was succeeded by many other journals: the American Journal of Computational Linguistics (1974–1978, 1980–1983), and then Computational Linguistics (1984–present). Since 1988, the journal has been published for the ACL by MIT Press. The annual meeting was first held in 1963 in conjunction with the Association for Computing Machinery National Conference. The annual meeting was, for a long time, relatively informal and did not publish anything longer than abstracts. By 1968, the society took on its current name, the Association for Computational Linguistics (ACL). The publication of the annual meeting's Proceedings of the ACL began in 1979 and gradually matured into its modern form. Many of the meetings were held in conjunction with the Linguistic Society of America, and a few with the American Society for Information Science and the Cognitive Science Society. The United States government sponsored much research from 1989 to 1994, characterized by an increase in author retention rates and an increase in research in some key topics, such as speech recognition, in ACL. By the 21st century, it was able to maintain authors at a high rate who coalesced in a more stable arrangement around individual research topics. In 1991, the group published a prototype for a text generator based on the universal grammar theory of Noam Chomsky. The system, nicknamed Parrot, relied on a finite set of syntactic transformations and a hand-curated lexicon. Despite some initial success, including experimentation with morpheme syntactics, funding halted after the research team encountered intractable difficulties with inflection and abstract locutions. == Annual Meeting of the ACL == Every year, the ACL holds the Annual Meeting of the ACL. The location lies in Europe in years zero modulo three, North America in years one modulo three, and Asia–Australia in years two modulo three. In 2020, the Annual Meeting received for the first time more submissions from China than the United States. == Activities == The ACL organizes several of the top conferences and workshops in the field of computational linguistics and natural language processing. These include: Annual Meeting of the Association for Computational Linguistics (ACL), the flagship conference of the organization Empirical Methods in Natural Language Processing (EMNLP) International Joint Conference on Natural Language Processing (IJCNLP), held jointly one of the other conferences on a rotating basis Conference on Computational Natural Language Learning (CoNLL) Lexical and Computational Semantics and Semantic Evaluation (SemEval) Joint Conference on Lexical and Computational Semantics (SEM) Workshop on Statistical Machine Translation (WMT) Besides conferences, the ACL also sponsors the journals Computational Linguistics and Transactions of the Association for Computational Linguistics (TACL). Papers and other presentations at ACL and ACL-affiliated venues are archived online in the open-access ACL Anthology. == Special Interest Groups == ACL has a large number of Special Interest Groups (SIGs), focusing on specific areas of natural language processing. Some current SIGs within ACL are: == Presidents == Each year, the ACL elects a distinguished computational linguist who becomes vice-president of the organization in the next calendar year and president one year later. Recent ACL presidents are:
FutureMedia
FutureMedia is a program that analyzes the state and future of digital, social, and mobile media. It functions as a collaborative initiative at Georgia Tech and the Georgia Tech Research Institute. FutureMedia consults approximately 500 faculty members working in those fields. == History == In 2019, Future Media expanded into the Direct-To-Consumer market by acquiring Australian watchmaker Oak & Jackal. == Programs == === FutureMedia Fest === The organization most recently hosted FutureMedia Fest 2010, a four-day conference (Oct 4–7, 2010) with a keynote addresses from Michael Jones, the chief technology advocate at Google. The event featured panels, workshops, and technology demonstrations. === FutureMedia Outlook === Contemporaneous with FutureMedia Fest 2010, the organization released the FutureMedia Outlook, an analysis of the future of media, concentrating on six major trends in those fields, including information overload, personalization, data integrity, an expectation of multimedia, augmented reality, and collaborative software.
MIDI Show Control
MIDI Show Control (MSC), is a real-time System Exclusive extension of the international Musical Instrument Digital Interface (MIDI) standard. MSC enables all types of entertainment equipment to communicate with each other through the process of show control. The MIDI Show Control protocol is a technical standard ratified by the MIDI Manufacturers Association in 1991, which allows entertainment control devices to talk with each other and with computers to perform show control functions in live and prerecorded entertainment applications. Just like musical MIDI, MSC does not transmit the actual show media - it simply transmits digital information about a multimedia performance. == How MSC works == When any cue is called by a user (typically a stage manager) and/or preprogrammed timeline in a show control software application, the show controller transmits one or more MSC messages from its 'MIDI Out' port. A typical MSC message sequence is: the user has just called a cue the cue is for lighting device 3 the cue is number 45.8 the cue is in cue list 7 MSC messages are serially transmitted in the same way as musical messages and are fully compatible with all conventional MIDI hardware; however, many modern MSC devices now use Ethernet communications for higher bandwidth and the flexibility afforded by networks. Other performance parameters are also transmitted, such as lighting desk submaster settings using MSC SET messages. All cues that a media control device is capable of playing are assigned MSC messages within the Show Controller's cue list and they are transmitted from its MIDI Out port at the appropriate show time, depending on the actions of the user and the show controller's internally timed sequences. All MSC-compatible instruments follow the MSC specification and thus transmit identical MSC messages for identical MSC events, such as the playing of a certain cue on the media controller. Since they follow a published standard, all MSC devices can communicate with and understand each other, as well as with computers that have been programmed to understand MSC messages using the MSC Command Set. All MSC compatible instruments have a built-in MIDI interface and many now follow one of the various MIDI-over-Ethernet protocols. == History == To create the MSC spec, Charlie Richmond headed the USITT MIDI Forum on their Callboard Network in 1990, which included developers and designers from the theatre sound and lighting industry from around the world. It is believed that this was the first international standard to be developed without a single physical meeting of the participants. This Forum created the MSC standard between January and September 1990. This was ratified by the MIDI Manufacturers Association (MMA) in January 1991, and the Japan MIDI Standards Committee (JMSC) later that year, becoming a part of the standard MIDI specification in August 1991. The first show to fully use the MSC specification was the Magic Kingdom Parade at Walt Disney World's Magic Kingdom in September 1991. == MIDI Show Control software ==
COVFEFE Act
The Communications Over Various Feeds Electronically for Engagement Act (COVFEFE Act), House Bill H.R. 2884, was introduced in the United States House of Representatives on June 12, 2017, during the 115th United States Congress. The bill was intended to amend the Presidential Records Act to preserve Twitter posts and other social media interactions of the President of the United States and require the National Archives to store such items. H.R. 2884 was assigned to the House Oversight and Reform Committee for consideration. While in committee, there were no roll call votes related to the bill. The bill died in committee. U.S. Representative Mike Quigley, Democrat of Illinois, introduced the legislation due to Donald Trump's routine use of Twitter, stating "In order to maintain public trust in government, elected officials must answer for what they do and say; this includes 140-character tweets. If the president is going to take to social media to make sudden public policy proclamations, we must ensure that these statements are documented and preserved for future reference". If enacted, the bill "would bar the prolifically tweeting president from deleting his posts, as he has sometimes done". The COVFEFE Act would have also treated a president's personal social media accounts (e.g., Trump's "@realDonaldTrump" Twitter account) the same as official social media accounts (e.g., the "@POTUS" Twitter account). == Background == The bill title refers to "covfefe", a word in a May 31, 2017 tweet that Trump sent at 12:06 AM EDT, reading "Despite the constant negative press covfefe". This incomplete tweet was liked and retweeted hundreds of thousands of times, making it one of the most popular tweets of 2017, as people speculated on its meaning. The tweet was deleted at 5:48 AM EDT. At 6:09 AM EDT, Trump's account tweeted "Who can figure out the true meaning of 'covfefe' ??? Enjoy!" During the May 31 White House press briefing, Hunter Walker of Yahoo! News asked White House press secretary Sean Spicer about the tweet and if there was any concern about the president sending out incoherent tweets that stay up for hours. Spicer responded, "I think the president and a small group of people know exactly what he meant" and offered no other explanation. This unexpected response spawned additional media attention and criticism for its cryptic meaning, with commentators unsure whether or not Spicer was joking. Callum Borchers of The Washington Post's The Fix noted that the Trump administration deliberately responded in a way that encouraged the media and the public to focus on covfefe instead of other controversies like the Russia investigation, resignation of White House communications director Michael Dubke, or U.S.-Germany relations. == Legal significance of Trump's tweeting == Trump's tweets have been legally significant in the past. White House Press Secretary Sean Spicer stated that Trump's tweets are "considered official statements by the President of the United States". Some of his tweets have contradicted his agenda by undercutting or contradicting statements of public officials as well as the arguments of U.S. Department of Justice attorneys seeking to defend Trump's decisions in court. A federal appellate court cited one of Trump's tweets in upholding a lower court's order blocking Trump's Executive Order 13780 from going into effect in 2017. Courts have been clear that Twitter statements can be used as evidence of intent. Before Trump's "@realDonaldTrump" Twitter account was suspended, he blocked a number of users, preventing them from viewing his tweets or posting public replies. A group associated with Columbia University filed a lawsuit on behalf of blocked users, called Knight First Amendment Institute v. Trump. Plaintiffs successfully argued that @realDonaldTrump reply threads constituted a "designated public forum" akin to a public meeting, and therefore blocking users based on their political viewpoints violated their constitutional right to freedom of speech. The Second Circuit upheld this ruling on July 9, 2019. Regardless of the failure of the bill, Trump's tweets have been archived in accordance with the Presidential and Federal Records Act Amendments of 2014.
Stochastic parrot
In machine learning, the term stochastic parrot is a metaphor that frames large language models as systems that statistically mimic text without real understanding. The word "stochastic" – from the ancient Greek "στοχαστικός" (stokhastikos, 'based on guesswork') – is a term from probability theory meaning "randomly determined". The word "parrot" refers to parrots' ability to mimic human speech. The term was introduced in a 2021 paper on AI ethics titled "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜" and authored by Timnit Gebru, Emily M. Bender, Angelina McMillan-Major, and Margaret Mitchell. The paper outlined possible risks associated with large language models (LLMs). In December 2020, it was the subject of a workplace dispute between Gebru (then co-leader of Google's Ethical Artificial Intelligence Team) and Google, which had requested the retraction of the paper. The incident culminated in Gebru's controversial departure from the company. The paper was later presented at the 2021 ACM Conference, and the term "stochastic parrot" has seen widespread use in academic research concerning generative AI and LLMs. The term has been interpreted negatively as an insult towards AI. == Background == Timnit Gebru is an AI ethics researcher, Emily M. Bender is a linguist specializing in computational linguistics, and Margaret Mitchell is a computer scientist specializing in algorithmic bias. Gebru had joined Google in 2018, where she co-led a team on the ethics of artificial intelligence with Mitchell. In late 2020, the paper "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜" was co-written by Gebru and five other researchers, four of whom were Google employees. The paper argues that large language models (LLMs) present significant risks such as environmental and financial costs, inscrutability leading to unknown dangerous biases, and potential for deception as LLMs do not understand the concepts underlying what they learn. The paper states that LLMs are "stitching together sequences of linguistic forms ... observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning." Therefore, they are labeled "stochastic parrots". === Dismissal of Gebru by Google === After the paper was submitted for consideration to the 2021 ACM Conference, Google requested that Gebru either retract the paper from the conference or remove the names of Google employees from it. Gebru refused to do so without further discussion, and emailed Google Research vice president Megan Kacholia that if the company could not explain the request for retraction and address other concerns regarding similar projects, she would plan to resign after a transition period, stating that they could "work on a last date". The following day, on December 2, 2020, Gebru received an email saying that Google was "accepting her resignation". Her abrupt firing sparked protests by Google employees and negative publicity for the company. == Usage == The phrase has been used by AI skeptics to signify that LLMs lack understanding of the meaning of their outputs. Sam Altman, CEO of OpenAI, used the term shortly after the release of ChatGPT in December 2022, tweeting "i am a stochastic parrot, and so r u". The term was nominated as the 2023 AI-related Word of the Year by the American Dialect Society. == Debate == Some LLMs, such as ChatGPT, have become capable of interacting with users in convincingly human-like conversations. The development of these new systems has deepened the discussion of the extent to which LLMs understand or are simply "parroting". According to machine learning researchers Lindholm, Wahlström, Lindsten, and Schön, the term "stochastic parrot" highlights two vital limitations of LLMs: LLMs are limited by the data they are trained on and are simply stochastically repeating contents of datasets. Because they are just making up outputs based on training data, LLMs do not understand if they are saying something incorrect or inappropriate. Lindholm et al. noted that, with poor quality datasets and other limitations, a learning machine might produce results that are "dangerously wrong". === Subjective experience === In the mind of a human being, words and language correspond to things one has experienced. For LLMs, according to proponents of the theory, words correspond only to other words and patterns of usage fed into their training data. Proponents of the idea of stochastic parrots thus conclude that statements about LLMs are due to "the human tendency to attribute meaning to text", and claim this occurs despite the LLMs not actually understanding language. === Fine-tuning === Kelsey Piper argued that the claim that LLMs are stochastic parrots or mere "next-token predictors" focuses on pre-training, ignoring that modern LLMs are also fine-tuned to follow instructions and to prefer accurate answers. === Hallucinations and mistakes === The tendency of LLMs to pass off false information as fact is held as support. Called hallucinations or confabulations, LLMs will occasionally synthesize information that matches some pattern. LLMs may fail to distinguish fact and fiction, which leads to the claim that they can't connect words to a comprehension of the world, as humans do. Furthermore, LLMs may fail to decipher complex or ambiguous grammar cases that rely on understanding the meaning of language. For example: The wet newspaper that fell down off the table is my favorite newspaper. But now that my favorite newspaper fired the editor I might not like reading it anymore. Can I replace 'my favorite newspaper' by 'the wet newspaper that fell down off the table' in the second sentence? GPT-4, an LLM released in March 2023, responded yes, not understanding that the meaning of "newspaper" is different in these two contexts; it is first an object and second an institution. === Benchmarks and experiments === One argument against the hypothesis that LLMs are stochastic parrot is their results on benchmarks for reasoning, common sense and language understanding. In 2023, some LLMs have shown good results on many language understanding tests, such as the Super General Language Understanding Evaluation (SuperGLUE). GPT-4 scored in the >90th-percentile on the Uniform Bar Examination and achieved 93% accuracy on the MATH benchmark of high-school Olympiad problems, results that exceed rote pattern-matching expectations. Such tests, and the smoothness of many LLM responses, help as many as 51% of AI professionals believe they can truly understand language with enough data, according to a 2022 survey. === Expert rebuttals === Some AI researchers dispute the notion that LLMs merely "parrot" their training data. Geoffrey Hinton, a pioneering figure in neural networks, counters that the metaphor misunderstands the prerequisite for accurate language prediction. He argues that "to predict the next word accurately, you have to understand the sentence", a view he presented on 60 Minutes in 2023. From this perspective, understanding is not an alternative to statistical prediction, but rather an emergent property required to perform it effectively at scale. Hinton also uses logical puzzles to demonstrate that LLMs actually understand language. A 2024 Scientific American investigation described a closed Berkeley workshop where state-of-the-art models solved novel tier-4 mathematics problems and produced coherent proofs, indicating reasoning abilities beyond memorization. The GPT-4 Technical Report showed human-level results on professional and academic exams (e.g., the Uniform Bar Exam and USMLE), challenging the "parrot" characterization. Anthropic conducted mechanistic interpretability research on Claude, using attribution graphs to identify circuits. The research showed how the LLM processes information via chains of fuzzy logical inference, and indicated an ability to plan ahead. They found that Claude 3.5 Haiku "employs remarkably general abstractions", forms "internally generated plans for its future outputs" and "works backwards from its longer-term goals". They noted that "The mechanisms of the model can apparently only be faithfully described using an overwhelmingly large causal graph." They also found that the model includes "mechanisms that could underlie a simple form of metacognition", in that it "thinks about" the level of its own knowledge before reaching its answer. === Interpretability === Another line of evidence against the 'stochastic parrot' claim comes from mechanistic interpretability, a research field dedicated to reverse-engineering LLMs to understand their internal workings. Rather than only observing the model's input-output behavior, these techniques probe the model's internal activations, which can be used to determine if they contain structured representations of the world. The goal is to investigate whether LLMs are merely manipulating surface statistics or if t
Software-defined mobile network
Software-defined mobile networking (SDMN) is an approach to the design of mobile networks where all protocol-specific features are implemented in software, maximizing the use of generic and commodity hardware and software in both the core network and radio access network (RAN). == History == Through the 20th century, telecommunications technology was driven by hardware development, with most functions implemented in special-purpose equipment. In the early 2000s, generally available CPUs became cheap enough to enable commercial software-defined radio (SDR) technology and softswitches. SDMN extends these trends into the design of mobile networks, moving nearly all network functions into software. The term "software-defined mobile network" first appeared in public literature in early 2014, used independently by Lime Microsystems and researchers from University of Oulu, Finland. == Limitations of hardware-based mobile networks == Mobile networks based on special-purpose hardware suffer from the following limitations: They have limited provisions for upgrades and usually must be replaced entirely when new standards are introduced. The individual components are not scalable in terms of performance and capacity, because the capacity of a component is fixed by the hardware implementation. Specialized equipment and its associated specialized software require vendor-specific training for the mobile operator's staff. Specialized hardware systems are usually supported and serviced by a single vendor, resulting in vendor lock-in. == Characteristics of SDMN designs == === Use of software-defined radio === SDR is an important element of SDMN, because it replaces protocol-specific radio hardware with protocol-agnostic digital transceivers. While many earlier digital radio systems used field-programmable gate arrays (FPGAs) or special-purposed digital signal processors (DSPs) for calculations on baseband radio waveforms, the SDMN approach moves all of the baseband processing into general-purpose CPUs. SDMN radio systems also use hardware with publicly-documented interfaces that is designed to be readily reproducible by multiple manufacturers. === Commodity components === SDMN designs avoid the use of components that are specialized as to their functions or that are available from only a single vendor. This is true of both the hardware and software elements of the network. === Software switching and transcoding === The telephony switches of SDMN networks are software-based, including software transcoding for speech codecs. === Centralized, distributed, or hybrid? === A new SDN architecture for wireless distribution systems (WDSs) is explored that eliminates the need for multi-hop flooding of route information and therefore enables WDNs to easily expand. The key idea is to split network control and data forwarding by using two separate frequency bands. The forwarding nodes and the SDN controller exchange link-state information and other network control signaling in one of the bands, while actual data forwarding takes place in the other band. == Advantages of SDMN == The SDMN approach has many advantages over hardware-based mobile network designs. Because SDMN hardware is protocol-agnostic, upgrades are software-only, even across technology generations. In the radio network, these changes can even be made on a site-by-site basis. Because SDMN hardware is designed to be easily sourced and reproduced: SDMN equipment can be serviced by a wider range of vendors, lowering maintenance costs. SDMN equipment can be manufactured anywhere in the world, lowering production costs. Because SDMN software is based on commodity operating systems and development tools: Support staff can be trained more quickly because they are already familiar with the underlying software systems. Many aspects of the SDMN can be monitored and managed with pre-existing tools, because they are already available in the commodity operating systems. Because SDMN network components run on general purpose computers, the network components can be scaled up in capacity by adding more computing power.