N1005 IBM Research: ibm research tokyo

Tired of answering questions? Let a smart machine help.

<div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;"></span></div><span lang="EN-US"><a href="http://researcher.ibm.com/researcher/view.php?person=jp-KAMIJOH"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Koichi Kamijo</span></a></span><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">, a scientist at IBM’s research lab in Tokyo was only 20 questions into a 300-question life insurance form. And he was already exhausted. There had to be an easier way to answer these questions. So, he got together with his colleagues, </span><span lang="EN-US"><a href="http://researcher.ibm.com/researcher/view.php?person=jp-RYOKAWA"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Ryo Kawahara</span></a></span><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">, </span><span lang="EN-US"><a href="http://researcher.ibm.com/researcher/view.php?person=jp-OSOGAMI"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Takayuki Osogami,</span></a></span><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;"> Masaki Ono and Shunichi Amano to come up with a cognitive tech answer to the cumbersome process of finding a life insurance product. And they identified two critical issues with the life insurance questionnaire that needed to be solved to make life plan simulator more user-friendly and useful for consumers and financial planners.</span> <br /><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt; mso-no-proof: yes;">                                                 </span><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;"></span></div><table align="center" cellpadding="0" cellspacing="0" class="tr-caption-container" style="margin-left: auto; margin-right: auto; text-align: center;"><tbody><tr><td style="text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgEJdPJ17uLKTHo82p2igD77P1CRS2ogNOQQml39MSEdiTSOcSRYLcL4FyO9OMtV52rTAHA-DgcdX091mt3-SbBwKX3pR8jLVvAu82IkCwOe98-nVBX9_533N1XUaUJU8HUBdQMDYR2s3_U/s1600/Tokyo_team.jpg" imageanchor="1" style="margin-left: auto; margin-right: auto;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgEJdPJ17uLKTHo82p2igD77P1CRS2ogNOQQml39MSEdiTSOcSRYLcL4FyO9OMtV52rTAHA-DgcdX091mt3-SbBwKX3pR8jLVvAu82IkCwOe98-nVBX9_533N1XUaUJU8HUBdQMDYR2s3_U/s1600/Tokyo_team.jpg" /></a></td></tr><tr><td class="tr-caption" style="text-align: center;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: x-small;">L-R: Masaki Ono, Ryo Kawahara, Koichi Kamijo, Takayuki Osogami<br />and Shunichi Amano of IBM Research – Tokyo</span></td></tr></tbody></table><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;"></span><br /><a name="more"></a><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Typically, consumers are asked to answer an average of 300 questions, from name, age, and marital status, to expected salary at the age of retirement, and education expenses for their children. The first issue Koichi and his team recognized: there were too many questions! Going through hundreds of questions is too much for people, even though they help financial planners understand their needs. The second issue he and his team realized they needed to solve was how to help financial planners also predict potential needs. With his colleagues’ expertise in machine learning, business analytics and natural language processing – along with ST-Life (Smile and Thanks Life), whose life plan simulator Dynamic Life Plan Navigator has been widely used by enterprises and organizations in Japan – they developed an insurance information lifecycle management platform to solve these critical issues. <span style="mso-spacerun: yes;"> </span></span><br /><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt; mso-no-proof: yes;">  </span><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;"></span></div><div class="separator" style="clear: both; text-align: center;"></div><div align="left" class="MsoNormal" style="text-align: left;"><div class="separator" style="clear: both; text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiMXS4y1A49dw4Jns2tN1pwRcqYpnGh7yj2EvZn3iFBProaIAKvGHEpqR6XVwVokMY72yS0gLt8OLl14Y6pZkIFROB9HfBUhO4vtMPQ1S6X_0RPcXVBYHcAv4RgVLU5Bdw9WO2IJJDsE72K/s1600/life_sim.jpg" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiMXS4y1A49dw4Jns2tN1pwRcqYpnGh7yj2EvZn3iFBProaIAKvGHEpqR6XVwVokMY72yS0gLt8OLl14Y6pZkIFROB9HfBUhO4vtMPQ1S6X_0RPcXVBYHcAv4RgVLU5Bdw9WO2IJJDsE72K/s1600/life_sim.jpg" /></a></div></div><div align="left" class="MsoNormal" style="text-align: left;"><br /> <br /><div align="left" class="MsoNormal" style="text-align: left;"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">First, Koichi and his team combined two components on the Dynamic Life Plan Navigator: its parameter estimator and human response simulator. The parameter estimator minimized the burden of answering 300 questions down to only eight by applying machine learning and analytics technologies. Just by answering questions like marital status, gender, number of children, income level, date of birth, and home mortgage status, the system then analyzes other users’ answers to automatically fill in answers to the other 292 questions. According to the recent experiment, people who successfully went through life plan simulation questions nearly doubled.</span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><table align="right" border="1" cellpadding="0" cellspacing="0" class="MsoTableGrid" style="border-collapse: collapse; border: none; margin-left: 6.75pt; margin-right: 6.75pt; mso-border-alt: solid white 1.5pt; mso-border-insideh-themecolor: background1; mso-border-insideh: 1.5pt solid white; mso-border-insidev-themecolor: background1; mso-border-insidev: 1.5pt solid white; mso-border-themecolor: background1; mso-padding-alt: 0in 5.4pt 0in 5.4pt; mso-table-anchor-horizontal: margin; mso-table-anchor-vertical: paragraph; mso-table-left: right; mso-table-lspace: 9.0pt; mso-table-overlap: never; mso-table-rspace: 9.0pt; mso-table-top: 2.85pt; mso-yfti-tbllook: 1184;"> <tbody><tr style="mso-yfti-firstrow: yes; mso-yfti-irow: 0; mso-yfti-lastrow: yes;">  <td style="border: solid white 1.5pt; mso-border-themecolor: background1; padding: 0in 5.4pt 0in 5.4pt; width: 2.25in;" valign="top" width="216"><div align="left" class="MsoNormal" style="mso-element-anchor-horizontal: margin; mso-element-anchor-vertical: paragraph; mso-element-frame-hspace: 9.0pt; mso-element-left: right; mso-element-top: 2.85pt; mso-element-wrap: around; mso-element: frame; mso-height-rule: exactly; text-align: left;"><b style="mso-bidi-font-weight: normal;"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">CREST   Grant</span></b></div><div align="left" class="MsoNormal" style="mso-element-anchor-horizontal: margin; mso-element-anchor-vertical: paragraph; mso-element-frame-hspace: 9.0pt; mso-element-left: right; mso-element-top: 2.85pt; mso-element-wrap: around; mso-element: frame; mso-height-rule: exactly; text-align: left;"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">The preference estimation uses   a technology that IBM Research -Tokyo has researched and developed with a grant   from the </span><a href="http://www.jst.go.jp/kisoken/crest/en/index.html"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Japan   Science and Technology Agency’s Core Research for Evolutionary Science and   Technology (CREST)</span></a><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">.</span></div></td> </tr></tbody></table><div align="left" class="MsoNormal" style="text-align: left;"><span style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Next, the human simulator can help discover a user’s potential preferences, such as insurance premium, coverage type, or what they consider important in choosing a product. The estimator creates several unconventional, two-option hypothetical questions based on the user’s data from the past. Then the simulator suggests an option. By repeating this process four to five times, the simulation system uncovers the user’s potential preferences, including the ones which the user has never imagined, such as job, partner and vacation destination preferences. </span></div></div><div align="left" class="MsoNormal" style="text-align: left;"><br /><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">More than 97 percent of the users who completed these hypothetical questions in the recent experiment did not feel it was a burden. And felt that they received an appropriate life insurance product recommendation.</span><b style="mso-bidi-font-weight: normal;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;"> </span></b><br /><div class="separator" style="clear: both; text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiOBxDlT-2xxmtYYIPxVNnJhwRMQp52Gkr-wut4hmJLCEiFoO9fRKdd-X3lTpXI2WbQROz9ATqg3LenKsyOgU2oXvvdjWlzRULv5dWgtchSZ8LoqxWXe3Qlpwnv8AIM58p0X2ZZc3-E8hQH/s1600/lifesim_question.jpg" imageanchor="1" style="margin-left: 1em; margin-right: 1em;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiOBxDlT-2xxmtYYIPxVNnJhwRMQp52Gkr-wut4hmJLCEiFoO9fRKdd-X3lTpXI2WbQROz9ATqg3LenKsyOgU2oXvvdjWlzRULv5dWgtchSZ8LoqxWXe3Qlpwnv8AIM58p0X2ZZc3-E8hQH/s1600/lifesim_question.jpg" /></a></div><b style="mso-bidi-font-weight: normal;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Protecting Your Life (Insurance) Choices</span></b></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Additionally, the research team created a bonus technology to help avoid posting inappropriate expressions and private information to online life insurance-related forums. In Japan, some users want to reference and reply to “people-like-me” comments in these forums that are often managed by volunteer life insurance agents and financial advisors. But to post a question can be a challenge because it can be difficult to determine what information to contribute, while not disclosing private information.</span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">The context-aware discussion forum technology the team built highlights potentially inappropriate expressions and private information to help users amend their posts before uploading them to a forum. Koichi and his team hope that it will also help prevent “flaming” (hostile interactions) in the forum due to improper comments. </span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><b style="mso-bidi-font-weight: normal;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Try, try again</span></b></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Koichi used the tool again to do his life plan, and this time, it only took him five minutes to answer eight questions, overwrite some past answers, and receive a detailed life insurance plan and product based on future income and stock market performance. </span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">Koichi and his team are now working on applying the machine learning platform to other industries, such as banks and education. For example, like life insurance products, banks may consider using this platform to offer products to their customers that suit their savings goals. In education, such a platform may be applied to help teachers understand their students’ preferences and abilities by asking questions of their parents. If teachers can anticipate each student’s preference and ability based on an estimated value the platform’s analytics technology provides, they may be able to come up with a customized curriculum that effectively provides student support.<span style="mso-spacerun: yes;">  </span><span style="mso-spacerun: yes;">  </span></span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "helvneue light for ibm" , sans-serif; font-size: 13.0pt;">ST-Life continues to work with Koichi and his team to advance the technology. At IBM Japan, Koichi and his team plan to collaborate with IBM’s consulting unit to deploy the insurance information lifecycle management platform to the insurance industry. <span style="mso-spacerun: yes;"> </span></span></div>

Koichi Kamijo , a scientist at IBM’s research lab in Tokyo was only 20 questions into a 300-question life insurance form. And he was already...

Mimicking Neurons With Math

2:39 AM N1000 0

<br /><table cellpadding="0" cellspacing="0" class="tr-caption-container" style="float: left; margin-right: 1em; text-align: left;"><tbody><tr><td style="text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiZMQ-iS6m2JCeKaVr8Vf9mWlgf40Z83T_rCNJePVTsJ1o0pB3GZevnUT_lNZsIZbK6qg_VQf0cblgjqYoc1SLaGE6AMm_ZsFfgGktQAzww6Wg2qFWJIDxld6FiD8SvbRXa5rYBxw4L15k/s1600/Osogami.JPG" imageanchor="1" style="clear: left; margin-bottom: 1em; margin-left: auto; margin-right: auto;"><img border="0" height="150" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiZMQ-iS6m2JCeKaVr8Vf9mWlgf40Z83T_rCNJePVTsJ1o0pB3GZevnUT_lNZsIZbK6qg_VQf0cblgjqYoc1SLaGE6AMm_ZsFfgGktQAzww6Wg2qFWJIDxld6FiD8SvbRXa5rYBxw4L15k/s200/Osogami.JPG" width="200" /></a></td></tr><tr><td class="tr-caption" style="text-align: center;">Dr. Takayuki Osogami</td></tr></tbody></table><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11pt;"></span></div><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%;">Artificial neural networks have long been studied with the hope of achieving a machine with the human capability to learn. Today’s attempts at artificial neural networks are built upon <a href="https://en.wikipedia.org/wiki/Hebbian_theory">Hebb’s rule</a>, which Dr. Donald Hebb proposed in 1949 as how neurons adjust the strength of their connections. Since Hebb, other “rules” of neural learning have been introduced to refine Hebb’s rule, such as <a href="https://en.wikipedia.org/wiki/Spike-timing-dependent_plasticity">spike-timing dependent plasticity (STDP)</a>. All of this helps us understand our brains, but makes developing artificial neurons even more challenging.</span><br /><div class="MsoNormal"><br /></div><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%;">A biological neural network is too complex to exactly map into an artificial neural network. But IBM mathematician <a href="http://researcher.ibm.com/researcher/view.php?person=jp-OSOGAMI">Takayuki Osogami</a> and his team at <a href="http://www.research.ibm.com/labs/tokyo/index.shtml">IBM Research-Tokyo</a>might have figured it out by developing artificial neurons that mathematically mimic STDP to learn words, images, and even music. Takayuki’s mathematical neurons form a new artificial neural network, creating <a href="http://www.nature.com/articles/srep14149">a dynamic Boltzmann machine (DyBM)</a> that can learn about information from multiple contexts through training.</span><br /><br /><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;">The team taught seven artificial neurons the word “SCIENCE” (one artificial neuron per bit) in a form of a bitmap image. So, the image of “SCIENCE” becomes:</span><br /><br /><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">00111001100111011110100010011001111</span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">01000010010010010000100010100101000</span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">01000010000010010000110010100001000</span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">00110010000010011100101010100001110</span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">00001010000010010000100110100001000</span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">00001010010010010000100010100101000</span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">01110001100111011110100010011001111.</span></div><div align="left" class="MsoNormal" style="text-align: left;"></div><br /><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%;">The “1s” equate to the lines making up the letters, while the “0s” translate to the white space around the letters.</span><br /><div class="MsoNormal"><br /></div><div class="MsoNormal"><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%;">What these seven neurons can do all at once is read and write 7-bit information. The word “SCIENCE” is expressed as 7-bit x 35-bit of pattern sequences that equals a 245 bits monochrome bitmap image. These seven neurons read and memorized each piece of the 7-bit information in the image. For example, “0100010” is the tenth column of the whole image according to the learning order, and the neurons recall those pieces in the order they learned them. By memorizing the word from left to right and from right to left, the neurons could recognize “SCIENCE” forward and backward, or in any order – like how we might solve a jumble word puzzle or crossword puzzle.</span><br /><br /></div><table align="center" cellpadding="0" cellspacing="0" class="tr-caption-container" style="margin-left: auto; margin-right: auto; text-align: center;"><tbody><tr><td style="text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEighV2abZtNNw-CS521a7d0GR5uxClotHXni2sRm8CgdSF-ODM8hOnNd-cENqMztWenHGPHtALCVLHRp0D4pLxQW-mhGCToDrFzYbg75oW_wEpI1d9K_iPiV_v0hVJEfrrFzU9TPAwJDmOL/s1600/science.jpg" imageanchor="1" style="margin-left: auto; margin-right: auto;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEighV2abZtNNw-CS521a7d0GR5uxClotHXni2sRm8CgdSF-ODM8hOnNd-cENqMztWenHGPHtALCVLHRp0D4pLxQW-mhGCToDrFzYbg75oW_wEpI1d9K_iPiV_v0hVJEfrrFzU9TPAwJDmOL/s1600/science.jpg" /></a></td></tr><tr><td class="tr-caption" style="text-align: center;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">Figure 1: The DyBM successfully learns two target sequences<br />and retrieves a particular sequence when an associated<br />cue is presented.<span style="mso-spacerun: yes;"> </span>(Credit: Scientific Reports)</span></td></tr></tbody></table><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;"></span><br /><div align="left" class="MsoNormal" style="text-align: left;"> </div><div align="left" class="MsoNormal" style="text-align: left;"></div><div align="left" class="MsoNormal" style="text-align: left;"><span style="mso-spacerun: yes;"><span style="color: red;"><span style="mso-spacerun: yes;"></span></span><b><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%;">More neurons. More memories.</span></b></span><br /><div class="MsoNormal"><br /><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%;">Things get more complicated (and interesting) when these artificial neurons learn about different topics in different formats, such as the human evolution image, below. Takayuki’s team put 20 artificial neurons to the task of learning this image, which shows how we humans have evolved, from left to right. Why they used 20 artificial neurons this time? One column of the image showing the human evolution consists of 20 bits, and they made it consistent with it.</span><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;"> </span></div><div class="MsoNormal"><br /><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;">These neurons learned how the pieces of the image line up in the correct order of evolution – from apes to Homo sapiens. As Takayuki runs simulation, these neurons learn more over time, while detecting the mistakes they make – and making corrections per simulation. With each simulation, the neurons generate an image to show their progress in re-creating the image. It took just 19 seconds for the 20 neurons to learn the image correctly, as mapped out below.</span> </div></div><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;"><span style="color: red;"><span style="mso-spacerun: yes;">    </span></span></span><br /><table align="center" cellpadding="0" cellspacing="0" class="tr-caption-container" style="margin-left: auto; margin-right: auto; text-align: center;"><tbody><tr><td style="text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhduMKWBbUD2ICeaGmvzZJVvo9rBCIrOK0br3poStGJgZ6o8ynRBzqhAIEvtQ5bXkRr2M8MqPansPzzaEDf5m7lGk7ssuDMlMa-N6ntb-cfTN9gKyS67RyECAkut_QsknM6NLmUdeGciVk/s1600/humanrevolution.png" imageanchor="1" style="margin-left: auto; margin-right: auto;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhduMKWBbUD2ICeaGmvzZJVvo9rBCIrOK0br3poStGJgZ6o8ynRBzqhAIEvtQ5bXkRr2M8MqPansPzzaEDf5m7lGk7ssuDMlMa-N6ntb-cfTN9gKyS67RyECAkut_QsknM6NLmUdeGciVk/s1600/humanrevolution.png" /></a></td></tr><tr><td class="tr-caption" style="text-align: center;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">Figure 2: The DyBM learned the target sequence of human evolution. (Credit: Scientific Reports)</span></td></tr></tbody></table><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;"></span></div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri; mso-fareast-language: JA; mso-no-proof: yes;"></span><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;"></span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-font-weight: bold; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;">Images and text are one thing. But neurons encompass all senses. So, Takayuki’s team put 12 of their artificial neurons to work learning music. Using a simplified version of the German folk song, <i>Ich bin ein Musikante</i>, each neuron was assigned to one of the 12 notes (Fa, So, Ra, Ti, Do, Re, Mi, Fa, So, Ra, Ti, Do). After 900,000 training sessions, they learned the sequential patterns of tones to the point of being able to generate a simplified version of the song.</span></div><div align="left" class="MsoNormal" style="text-align: left;"><br /><div class="separator" style="clear: both; text-align: center;"><iframe allowfullscreen="" class="YOUTUBE-iframe-video" data-thumbnail-src="https://i.ytimg.com/vi/vCcR3ii9FkA/0.jpg" frameborder="0" height="266" src="https://www.youtube.com/embed/vCcR3ii9FkA?feature=player_embedded" width="320"></iframe></div><br /></div><div align="left" class="MsoNormal" style="text-align: left;"><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-font-weight: bold; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;">The neurons learn music much like we might: repetition from beginning to end to the point of memorization. Currently, the 12 neurons only generate quarter notes, but simply doubling the neurons to 24 gives the system the ability to comprehend half-notes.</span></div><table align="center" cellpadding="0" cellspacing="0" class="tr-caption-container" style="margin-left: auto; margin-right: auto; text-align: center;"><tbody><tr><td style="text-align: center;"><a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjg4U6zhiFcBQEiy5fe8MUJxdvj68d0SV6yz6QGmWzfZLEVWXOO3l8TNgQw4qcJAhtLbTQFWYURqikz5TXJJuxmF5vAx9aUA7_dWxg5W6yye8qQZtH3o_mkvi_V1mrFzojUCy7Lt_dXyDI/s1600/music.png" imageanchor="1" style="margin-left: auto; margin-right: auto;"><img border="0" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjg4U6zhiFcBQEiy5fe8MUJxdvj68d0SV6yz6QGmWzfZLEVWXOO3l8TNgQw4qcJAhtLbTQFWYURqikz5TXJJuxmF5vAx9aUA7_dWxg5W6yye8qQZtH3o_mkvi_V1mrFzojUCy7Lt_dXyDI/s1600/music.png" /></a></td></tr><tr><td class="tr-caption" style="text-align: center;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;">Figure 3: The DyBM learned the target music. (Credit: Scientific Reports)</span></td></tr></tbody></table><div align="left" class="MsoNormal" style="text-align: left;">   </div><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri; mso-fareast-language: JA; mso-no-proof: yes;"></span><span lang="EN-US" style="font-family: "Calibri",sans-serif; font-size: 11.0pt; mso-bidi-font-family: Calibri;"></span></div> <br /><div class="MsoNormal"><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-bidi-font-weight: bold;">Takayuki’s DyBM not only memorizes and recalls sequential patterns, but it can also detect anomalies in sequential patterns, and make predictions about future patterns. It could be used to predict driving risks through car-mounted cameras, or generate new music, or even detect and correct grammatical errors in text. Takayuki’s work, currently funded by the Japan Science and Technology Agency’s Core Research for Evolutionary Science and Technology (CREST), hopes to advance his DyBM by integrating it with reinforcement learning techniques to optimally act on the basis of such anomaly detection and prediction.</span></div><div class="MsoNormal"><br /></div><span style="font-family: "HelvNeue Light for IBM",sans-serif; font-size: 13.0pt; line-height: 107%; mso-ansi-language: EN-US; mso-bidi-font-family: "Times New Roman"; mso-bidi-font-weight: bold; mso-bidi-language: AR-SA; mso-fareast-font-family: Calibri; mso-fareast-language: EN-US;">The scientific paper <a href="http://www.nature.com/articles/srep14149"><i>Seven neurons memorizingsequences of alphabetical images via spike-timing dependent plasticity</i></a> by <a href="http://researcher.ibm.com/researcher/view.php?person=jp-OSOGAMI">Takayuki Osogami</a> and <a href="https://sites.google.com/site/motsukapublic/">Makoto Otsuka</a> appears in Scientific Reports of the Nature Publishing Group on September 16, 2015, DOI: 10.1038/SREP14149.</span><br /><div align="left" class="MsoNormal" style="text-align: left;"><span lang="EN-US"></span></div><div align="left" class="MsoNormal" style="text-align: left;"></div>

artificial intelligence artificial neurons hebb's rule ibm research tokyo Mimicking Neurons With Math

Dr. Takayuki Osogami Artificial neural networks have long been studied with the hope of achieving a machine with the human capability to lea...