The born of Hierarchical Temporal Memory: On Intelligence : Jeff Hawkins Sandra Blakeslee, 2004 A Hierarchical Bayesian Model of Invariant PatternRecognition in the Visual Cortex : Jeff Hawkins, Dileep George, 2004 Invariant Pattern Recognition using BayesianInference on Hierarchical Sequences :Jeff Hawkins, Dileep George, 2004 How the brain might work: A hierarchical and temporal model for learning and recognition : George Dileep, 2008 Numenta Publications: Hierarchical Temporal Memory: Concepts, Theory and Terminology : Jeff Hawkins, George Dileep, 2006 Numenta Node Algorithms Guide : Numenta Inc., 2007 HTM Comparison with existing models : Numenta Inc., 2007 Problems that fit HTM : Numenta Inc., 2007 Getting Started With NuPIC : Numenta Inc., 2008 Numenta Platform forIntelligent Computing: Node Plugin Developer’s Guide : Numenta Inc., 2008 Advanced NuPIC Programming : Numenta Inc., 2008 Numenta Vision Toolkit Tutorial : Numenta Inc. , 2009 Hierarchical Temporal Memory including HTM Cortical Learning Algorithms : Numenta Inc. , 2010 Other Publications based on HTM basics: Analysis and Implementation of theMemory - Prediction Framework : S.Garalevicius, 2005 Unsupervised Phoneme Aqusition using Hierarchical Temporal Models , L. Lee Majure, 2006 Memory–Prediction Framework for Pattern Recognition:Performance and Suitability of the Bayesian Model of Visual Cortex : S. Garalevicius, 2007 Online temporal pattern learning : N. Farahmand, M.H. Dezfoulian, H. GhiasiRad, A. Mokhtari, A. Nouri, 2009 Spatio–Temporal Memories for Machine Learning:A Long-Term Memory Organization : J. A. Starzyk, 2009 Hierarchical Bayesian Reservoir Memory : A. Nouri, H. Nikmehr, 2009 Works and Publications based on HTM: Hierarchical Temporal Memory Networks for Spoken Digit Recognition: J. van Doremalen, 2007 Handwritten Digit Recognition using HierarchicalTemporal Memory :B. Bobier, 2007 The Application of Hierarchical Temporal Memoryto the Evaluation of EEG Signals :J. M. Casarella, 2007 Song Identification using Numenta platform for intelligent computing : N. C. Schey, 2008 Spoken Language Identification With Hierarchical Temporal Memories : D. Robinson, K. Leung, X. Falco, 2009 Support for the use of hierarchical temporal memory systems in automated design evaluation: A first Experiment : J. Hartung, J. McCormack, F. Jacobus, 2009 On the Optimum Architecture of the Biologically Inspired Hierarchical Temporal Memory Model Applied to the Hand-Written Digit Recognition: S Štolc, I. Bajla, 2010 THIS ARTICLE FROM s.m.mohammadzadeh weblog I'm writing a report about using Hierarchical Temporal Memory to model kids behaviourlearning a second Language. I have Googled many times to find related works. But I noticed that there are just some works related to the HTM. I’ll upload them all here to have a quick reference. I didn't put link to the original materials to have always a copy of the originals and to be affected by web-site changes. Take note that some of the uploaded contents (in special Numenta Inc. published articles) are licensed and must be used according to the respective License.
一觉醒来，没想到摘抄的津巴多讲座竟然被加精了，跟武老师的讨论竟然也被加精了。 跟武老师讨论提到的直觉。我觉得需要稍微写几句。作为一个被外界认为只有直觉思维的学科，景观设计一直都很缺乏研究，学科研究能力很弱。所以我的论文希望知道如何通过科研，通过创造知识，建立景观专业的权威。第一步就是了解现在本学科的实践情况，知识结构，以及知识的运用情况。 我的博士论文结论有三部份数据， 第一部分，在景观领域，设计师和教育者都强调要多做研究，那么究竟我们自己认为什么算的上”研究“。 第二部分，调查设计师的思考过程中直觉和逻辑两种思考的运用情况，以及包括研究在内的知识的运用情况。 第三部分，调查具体不同研究论题在实践中被运用的情况，教育者的研究情况，以及设计师对未来知识的需求情况。讨论这些论题运用了什么策略来构建知识，来 建立景观专业的权威。 直觉是一种思维方式。是人们在不自觉的条件下收集长期记忆，并通过感觉表征出来的一种思维。这个有一本最近的畅销书（美国网上都买断了，我从英国买了一本）叫thinking, fast and slow。讲的这两种思维机制，特别是直觉思维。作者是诺贝尔经济学奖得主Daniel Kahneman，他主要研究决策和思维。得诺贝尔奖的是理性人假设误区的期望曲线。 哲学上很早就有对直觉思维的讨论，没有脑科学之前，他们真的很skewed up. 这是一个研究直觉哲学家的在TED自嘲： 我去开一个**会，人们听说我是大学教授，都对我投来同情的目光。后来我去开一个教育大会， 人们听说我是搞哲学的，也对我投来同情的目光。再 后来我去开一个哲学大会， 人们听说我是搞潜意识的，继续对我投来同情的目光。 直到脑科学测量出declarative memory 和non-declarative memory在大脑中由不同部位被处理： Kandel, E. R., Schwartz, J. H., Jessell, T. M., Siegelbaum, S. A., Hudspeth, A. J. (2013). Principles of neural science (5th ed.). New York: McGraw-Hill Medical. 我相信21世纪是神经认知学的世纪，是脑科学的世纪，也是人们重新认识直觉，重新理解人文学科的世纪。
http://scienceblog.com/ Researchers discover potential explanation for why a diet high in DHA improves memory Understanding What’s Up With the Higgs Boson Debilitating eyesight problems on decline for older Americans All-carbon solar cell harnesses infrared light Communication scheme makes popular applications ‘gracefully mobile’ Picatinny Army engineers set phasers to ‘fry’ After child dies, mom’s risk of early death skyrockets Success of fertility treatment may approach natural birth rate Probing the roots of depression by tracking serotonin regulation at a new level
Onyx: A Prototype Phase Change Memory Storage Array --------------------------------------------------- 摘要： We describe a prototype high-performance solid-state drive based on first-generation phase-change memory (PCM) devices called Onyx. Onyx has a capacity of 10 GB and connects to the host system via PCIe. We describe the internal architecture of Onyx including the PCM memory modules we constructed and the FPGA-based controller that manages them. Onyx can perform a 4 KB random read in 38 μs and sustain 191K 4 KB read IO operations per second. A 4 KB write requires 179 μs. We describe our experience tuning the Onyx system to reduce the cost of wear-leveling and increase performance. We find that Onyx out-performs a state-of-the-art flash-based SSD for small writes ( 2 KB) by between 72 and 120% and for reads of all sizes. In addition, Onyx incurs 20-51% less CPU overhead per IOP for small requests. Combined, our results demonstrate that even first-generation PCM SSDs can out-perform flash-based arrays for the irregular (and frequently read-dominated) access patterns that define many of today's "killer" storage applications. Next generation PCM devices will widen the performance gap further and set the stage for PCM becoming a serious flash competitor in many applications. 该文描述了基于第一代 PCM 设备的高性能 SSD 原型 Onyx 。该原型通过 PCIE 连接，具有 10GB 容量。给出了其内部结构，包括 PCM 模块和基于 FPGA 的控制器。 性能： 一次 4 KB 随机读 38μs ； 191K 次 4 KB 读操作 每秒；一次 4 KB 写 179 μs 。 实验描述了如何降低 wear-leveling 开销，增加性能； 发现：该原型性能超过当前的基于 flash 的 SSD ——小写（ 2KB ）性能： 72%-120% ，各种大小的读操作性能。同时，对于小请求发生小于 20-25%CPU 开销。 在非规则访问模式下，第一代的 PCM SSD 性能仍超过基于 Flash 的阵列。 结论：下一代 PCM 继续扩大该性能差异，成为 flash 的取代者。 注 1 ： PCM 相变存储器（ PCM ）是一种非易失存储设备，它利用材料的可逆转的相变来存储信息。同一物质可以在诸如固体、液体、气体、冷凝物和等离子体等状态下存在，这些状态都称为相。相变存储器便是利用特殊材料在不同相间的电阻差异进行工作的。 该论文描述了 PCM 存储设计的特点，相对于 flash SSD 。 参考： http://www.dzsc.com/data/html/2009-11-13/80113.html http://www.docin.com/p-87626039.html 注 2 ： Wear-leveling 关于 SSD 的寿命， MLC 芯片的参数是写一万次，虽然看上去有点小，不过主控芯片会使用一种叫 Wear Leveling( 磨损平衡 ) 的机制，就是把写分散在每个块上，来延长寿命。 Wear leveling 可保证平均使用 FLASH 闪存内的每一个区域的一种软件技术。使用该技术的好处在于：可以让 FLASH 闪存芯片的寿命更长，出错率更低。具体有动态模式和静态模式。
机器翻译【Machine translation (MT) 】和翻译记忆【Translation Memory (TM)】相结合，是一种相当有效的计算机辅助翻译方式。 因此，对这两方面的研究进展以及实际应用的情况都值得关注。 附件1 Machine translation (MT) is the application of computers to the task of translating texts from one natural language to another. One of the very earliest pursuits in computer science, MT has proved to be an elusive goal, but today a number of systems are available which produce output which, if not perfect, is of sufficient quality to be useful in a number of specific domains." A definition from the European Association for Machine Translation (EAMT), "an organization that serves the growing community of people interested in MT and translation tools, including users, developers, and researchers of this increasingly viable technology." http://www.aaai.org/AITopics/pmwiki/pmwiki.php/AITopics/MachineTranslation http://en.wikipedia.org/wiki/Machine_translation Language and Machines: Computers in Translation and Linguistics (1966) http://blog.sina.com.cn/s/blog_65197d930100rtlc.html Statistical Machine Translation http://www.statmt.org/ This website is dedicated to research in statistical machine translation, i.e. the translation of text from one human language to another by a computer that learned how to translate from vast amounts of translated text. Introduction to Statistical MT Research The Mathematics of Statistical Machine Translation by Brown, Della Petra, Della Pietra, and Mercer Statistical MT Handbook by Kevin Knight SMT Tutorial (2003) by Kevin Knight and Philipp Koehn ESSLLI Summer Course on SMT (2005), day1 , 2 , 3 , 4 , 5 by Chris Callison-Burch and Philipp Koehn. MT Archive by John Hutchins, electronic repository and bibliography of articles, books and papers on topics in machine translation and computer-based translation tools Conferences and Workshops See comprehensive list of NLP meetings . Software Giza++ a training tool for IBM Model 1-5 (version for gcc-4) Moses , a complete SMT system Pharaoh a decoder for phrase-based SMT Rewrite a decoder for IBM Model 4 BLEU scoring tool for machine translation evaluation Parallel Corpora LDC Linguistic Data Consortium Canadian Hansards Europarl Acquis Communitaire ELRA 附件2 A translation memory is a linguistic database that continually captures your translations as you work for future use. All previous translations are accumulated within the translation memory (in source and target language pairs called translation units) and reused so that you never have to translate the same sentence twice. The more you build up your translation memory, the faster you can translate subsequent translations, enabling you to take on more projects and increase your revenue. http://www.translationzone.com/en/translator-solutions/translation-memory/ http://en.wikipedia.org/wiki/Translation_memory The concept of a translation memory has been around for a long time--more than twenty years--but only recently has it become a significant commercial entity. Basically, a translation memory is a system which scans a source text and tries to match strings (a sentence or part thereof) against a database of paired source and target language strings with the aim of reusing previously translated materials. Some translation memories attempt only literal matching, ie can only retrieve the exact match of a sentence, while others employ fuzzy matching algorithms to retrieve similar target language strings, flagging differences. The flexibility and robustness of the matching algorithm largely determine the performance of the system, although for some applications (ie, highly repetitive material) the recall rate of exact matches can be high enough to justify the literal approach. Translation memories are typically integrated into translation workstation packages, where they can be used in tandem with a terminology management system, a multilingual dictionary, and even raw MT output. http://www.issco.unige.ch/en/research/projects/ewg95//node149.html
F1000 推荐了 PNAS 上的一篇 paper ： Exercise training increases size of hippocampus and improves memory 大脑海马发育完全后，人类进入老年期后，海马以每年 1%-2% 的比率缩小，也许这是很多精神病与老年痴呆症的主要原因。本研究证明：适当的锻炼身体可以增加大脑海马的体积，并能改善空间记忆。而且，更为重要的是，老年锻炼可以逆转缩小的海马，甚至引起海马的二次发育。 This is the first study to show that moderate physical training, even starting at an older age can remedy the hippocampal volume loss associated with normal aging. Healthy aging is associated with shrinkage of the hippocampal volume at about 1-2% per year, representing a risk factor for the development of cognitive impairment in older subjects. Due to the increasing number of people reaching old age, the question of how this physiological decline can be delayed is of fundamental social and medical interest. This study by Erickson and co-workers reveals an important contribution to this topic by showing that physical exercise is associated with an increase in hippocampal volume, possibly supporting improvement in spatial memory. The most interesting finding is that age-related loss of hippocampal volume can be remedied even when subjects start their physical training at older age. The authors investigated a group of 120 healthy subjects aged between 55 and 80 years. One group performed a moderate weekly walking training exercise. The second group took part in a stretching and toning programme for the same amount of time. Besides other measures, spatial memory and volume of the hippocampus, the caudate nucleus and the thalamus was investigated before training, six months and one year after starting the training. The analysis reveals a decrease of anterior hippocampal volume in the stretching group suggesting normal age-related neuronal loss. In contrast to this, the volume of the anterior hippocampus was increased by 2%, indicating that moderate physical training starting at older age yields a decrease in age-related neuronal degeneration. Interestingly, spatial memory was improved in both groups after one year of training. But, a direct relation between hippocampal volume and memory accomplishment was found in the walking group only. All in all, the results by Erickson and others strongly support the significance of physical fitness for cognitive performance. Moreover, the data indicate the benefit of physical training, even starting at an older age. Our conclusion: let’s get physical!
cudaThreadSynchronize() is a _host_ function that waits for all previous async operations (i.e. kernel calls, async memory copies) to complete. __synchtreads() is a _device_ function that acts as a thread barrier. All threads in a block must reach the barrier before any can continue execution. It is only of use when you need to avoid race conditions when threads in a block access shared memory.
评论：当前的技术无法解决DRAM和FLASM MEMORY在一个IC上共存的问题，但很多研究团队正在开展这方面研究，这个新闻介绍了NCSU小组的激动人心的研究进展。 A Flash Memory That Doubles as DRAM Engineers work to combine volatile and nonvolatile memory into one flash device By Joseph Calamia/March 2011 Photo: North Carolina State University 3 March 2011—Engineers at North Carolina State University (NCSU) have refurbished flash memory in an attempt to create something new: a "unified memory" type that can be fast but volatile, like the memory workhorse dynamic RAM, or slow but nonvolatile, like the flash storage in MP3 players. At this point, the team has simulated only one memory cell’s behavior and made a proof-of-concept prototype, but they believe that their design may lead to instant-on computers and power savings in today’s behemoth data centers. Of course, IEEE Spectrum readers have heard such memory claims before. More exotic memory technologies— resistive , phase change , even spintronic devices —are also contenders for the guts of imagined instant-on machines. The new flash’s advantage, says one of the NCSU designers, Daniel Schinke, is that it’s less adventurous: "Our device is also new, but the technology behind it is very mature." Traditional flash works by forcing charge onto a layer of metal or polycrystalline silicon called a floating gate. In terms of bits, charge on the gate represents a 1, and an absence of charge stands for a 0. A barricade of dielectrics surrounding the gate keeps the charge from escaping, even when the memory has no power. Alternatively, DRAM is much faster, keeping electrons in capacitors that charge quickly but need energy to keep their state. The NCSU flash has two floating gates instead of one. Storing all of its charge on the bottom gate, the flash can act like its old nonvolatile self. But by using the second gate and a continuous source of power, it can work more quickly, shuffling preset proportions of charge between each of the gates to represent a 1 or a 0. Tuo-Hung Hou, a professor of electronics engineering at National Chiao Tung University, in Taiwan, who did not work on the NCSU device but who also researches modified flash memories , agrees with Schinke. "The proposed is attractive," he says, "because it is based on very scalable, cheap, and production-proven technology." But to realize the promised applications, he says, the team will first need to prove that the design can perform as predicted in simulations, and he believes that getting it to operate in both modes without interference "might be challenging." Part of that challenge will be repurposing memory that is now optimized mainly to keep its content, not to endure many rewrite cycles, as DRAM is. The problem with traditional flash is that it requires a strong electric field set up in the dielectric to pull the charges from an electron channel onto the floating gate. Over many write cycles, that field can wear the dielectric down, making it easier for the electrons to escape. After some 100 000 rewrites, flash’s capabilities decline, but DRAM can survive though 10 orders of magnitude more cycles. The NCSU researchers, led by professor of electrical and computer engineering Paul Franzon and colleague Neil Di Spigna, think that using two closely neighbored floating gates could take the burden off the dielectric and make for faster memory in volatile mode. Between the channel and the gates, the device would have a thick layer of dielectric, as traditional flash does. But the dielectric layer between the two gates would be much thinner—thin enough to allow electrons to use direct tunneling from one gate to the next, a special form of quantum tunneling that can better preserve the dielectric. For dynamic programming, charge has to move only between the two floating gates. "We’re using a gentle mechanism," Franzon says, as opposed to the strong electric fields used for more "harsh" charging in traditional flash. That type of charging uses either a messier version of tunneling or forces the electrons to travel inside the dielectric. Tunneling the electrons more gently between gates might allow the large number of rewrite cycles found in more expensive DRAM, Franzon says. In nonvolatile mode, the device would pull charge from the electron channel using traditional means, cramming as much of it as possible into the lower, better-barricaded gate. In volatile mode it would use a constant supply of energy to quickly divvy the charge between the two gates and hold it there using electric fields. Before the power is turned off, Franzon says, the state in volatile mode could be stored in the lower gate—releasing all the charge for a 0 or giving it all back to the first gate for a 1. "In milliseconds you could transfer an entire memory…and freeze its contents," Franzon says. At start-up, that would mean less travel time for the processor to begin again, and thus a near instant boot. The ability to freeze data would also provide energy savings. Instead of powering volatile DRAM while it simply maintains the same memory, the double-gate device could temporarily store the information and cut the power until the data is really needed. Although Albert Fazio, an Intel Fellow and the company’s director of memory technology development, encourages the team to aim for such research goals, he cautions that the dual-gate device has a way to go before implementation. "There are constraints in this approach," he says, "that still leave it fairly deep in the research field." One difficulty he foresees is proving that the device can work in the dual modes in a real memory array configuration, not just as an individual memory cell. "It’s really the array configuration that tells you about the performance and also the cost," he says, and perhaps, he adds, whether it’s even worth unifying memory types. "If it’s more expensive than DRAM," Fazio asks, "well, then is it any better than using a discrete DRAM and a discrete flash?" TAGS: DRAM // flash // unified memory 来源： http://spectrum.ieee.org/semiconductors/memory/a-flash-memory-that-doubles-as-dram/?utm_source=techalertutm_medium=emailutm_campaign=031011
The Cyclone III M9K embedded memory blocks may exhibit bit error in which the read bit is a 1 when the expected bit is a 0. The problem is caused by bitline coupling in the read output. The issue is rare and requires the presence of multiple conditions for the M9K block to be susceptible to the bit error. The conditions include the use model of the M9K block, the application data pattern, and the operating conditions.Designs using the M9K blocks in dual clock and widest data width (x32 or x36) modes are most susceptible to the bit error. Designs using the M9K blocks in single clock or narrower data width modes are not affected when operating within data sheet specifications. The problem is highly data-pattern dependent and triggered by specific data bit combinations. Lastly, the problem can be exacerbated by lower temperature and lower voltage operations. The presence of all these conditions does not imply a bit error would necessarily occur. In addition, if some or all of the conditions are not present, the error will not occur.
Figuration of English Words in Outlook and Sound Introduction The idea had been in my brain for more than 20 years to example words before I published it on my blog on 22th Sept. 2007. The list of letter meanings is the central body of the idea. The 26 letters each has its own meaning based on its outlook or on a derivation summarized from thousands of words. However, the meaning of a letter shows flexibility, and letter O, for example, looks not only a round blur puzzle but also a circle area and a ring to link two parts into a new word as well. We need the flexibility to figure out the meaning of million words individually. The list also shows the meanings of only some two-letter groups because meanings of the other groups as word have been defined in common dictionaries or as word fixes in textbooks. The meaning of each group has a typical way of flexibility: one comes from the list and the other is comes from the meanings of its constitutional letters. Cu, for example, means “cumulate” in the list. Cu can also mean “cut down” because C means “cut” and U means “down” in the list. It is generally not necessary to define meanings of any letter groups having more than three letters because you can find their meanings out of a dictionary. The idea is that you may like to hold a copy of the list in one of your hands to figure out meaning of any a word in your own story of figuration. After some days you will not need the list any more except a few occasional checks because the list is easy to remember. You may need to know that meaning of consonant letters plays a more important role than the vowels and a vowel letter usually behaves like an emphasis to the meaning of the consonant or consonants in front of it without much its own meaning, i.e. a vowel with its preceding consonant or consonants usually makes only one meaning. A very important skill is to break a word into letter groups. A consonant with its next following vowel should be a group but how to break connected consonants and vowels, and even to separate sub-word groups are optional. There are no rules but skills. Sometimes, you need to add or to take off a letter in letter groups to explain a word because people did so when they create words for shortness or good look. For example, dispirit=dis+spirit; distress=dis+stress; account=a+count and applause=a+plause. Let us begin with the word “love”. Why is “love” consisted of these four specific letters? Why love means to load venom or change to your heart or to others? Is it the real story that the word “love” was invented many many years ago in that way? It does not matter but the figuration may be interesting and helpful to remember it. Do you believe that the idea is a miracle? You can find that the idea and the list work well for thousands and thousands words. It is even true that your stories of figuration for many words are exactly the same as they are in the textbooks of word origin in prefixes, suffixes and stems from Greek, Latin and European colloquies including old English or in a text book of etymology. The principle is that any word came to its present meaning in its reason of formation or figuration and shall we find it? The history lost the figuration and shall we discuss and make an agreement to define it now? The idea is intended to supply people a way to remember words easily and funnily. However, who cares about your stories of the figuration being true or not in history of word origin provided they help you to remember words and they are interesting. Your word stories may not the same as mine or you may be surprised to find out yours are exactly the same of mine. However most importantly, if you share them with people, you may find that you are special and genius. If you share the stories of figuration from one of your friends to compare with yours, you may find out that your friend is special and he/she is a friend of yours in sake of nature and personality. To exchange stories of a specific word or a group of words would be an interesting game in an intimate circle of friends confidentially. You will remember words in your own best way by the figuration learning and practice. A word looks no longer hard and tedious but active, affective and interesting. Every one will become a writer to author a personal word storybook like a diary book and it may be published soon. If you would like to join my work, especially a native English speaker and a language specialist, if you would like to help me or join me to publish a book of the word stories in a form of a dictionary, would you please feel free to contact me by email: ypzong@mail.neu.edu.cn or feed back on my blog? Thank you!
标签: memory
