J288332 九、發明說明： 【發明所屬之技術領域】 本發明係屬於一種分析辨識 智慧分析辨識預測方法，可為一種人工 一電子計算機硬體，以藉由該電=體μ而安裝於 該方法可分析及辨識所輸入的,體執仃功能， 【先前技術】 爿並传到一預測結果。 :面上已存在有多種辨識分析的軟硬體，例 識裝置’可用於核對輸入的指紋是否存在於裝置中内建： 資料庫，又例如手寫辨識軟體，可分析 建的 否吻合於内建字庫的字體。 ‘，，、刑的予體是 目前的分析辨識軟硬體僅能以内建的數學 行所輸入目標資料的辨識及分析，不但存在—定㈣誤 率，其分析模式具有偏差，& Α + 曰' ，偈圭且無法經由錯誤來自行修正辨 識的規則及模式，來提高辨識及分析的精確度，此外，現 有的分析辨識軟硬邀對事物複雜的高維關係之行為模式， ，能以低維度統計模型或人為設定模型權值，造成不精密 完整的分析，以及不正確的判斷。 【發明内容】 故本發明人根據目前市面上各種分析辨識軟硬體缺乏 自行修正辨識分析規則的缺點，進而發明出一種人工智慧 分析辨識預測方法。 ’ 本發明之主要目的，係提供一種人工智慧分析辨識預 測方法，來建立一套智慧型的辨識、分析及預測準則，分 (Ε ^288332 析辨識一預測目扭 軚，並給予一預測結果，接著比 的結果與預測結果，修正的、 測的精確度。 —辨識規則，提高預 為達上述目的，本發明係令前述人JL知攀八批 體内，以成為1硬體型態安裝在一電子計算機硬 下步驟： 工智慧线來執行其錢，且包含有以 接爻複數筆的輪入資料，其中各筆資料含有溢 徵值及-個_ ; 聿貝^有複數個特 以一内建的機器學習演算法學習所輸入資料； 根據機器學習演算法的學習結果，建立 識一預測目標之模次盥招目丨^ ^ 刀析及辨 :::響預測目標結果變異之因素群及影響：I: 別擁有的影響權重值； I厅刀 接文一預測目標之輸入資料，1 特徵值； -中該貝枓包含複數個 ~以該模式與規則來辨識與分析該預測目標資料，並推 4于該目標之類別； 輪入該目標於事實上之類別 光μ 貝您類W並比對该事實類別與前 ^隹彳于之類別，若兩者比對相同則結束本方法，若比對不 同’則進行下列複數個學習步驟： 料根據該預測目標的資料及事實上的類別產生一筆新資J288332 IX. Description of the invention: [Technical field of the invention] The present invention belongs to an analysis and identification intelligent analysis identification prediction method, which can be an artificial electronic computer hardware, and the method can be installed by the electric body μ. Analyze and identify the input, body function, [previous technique] and pass a prediction result. There are many kinds of software and hardware for identification and analysis on the surface. The example device can be used to check whether the input fingerprint exists in the device: data library, for example, handwriting recognition software, can be analyzed whether the built-in is consistent with the built-in The font of the font. ',,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,曰' , 偈 且 and can not correct the rules and patterns of identification through mistakes to improve the accuracy of identification and analysis. In addition, the existing analysis identifies soft and hard behaviors that appeal to complex high-dimensional relationships Low-dimensional statistical models or artificially set model weights, resulting in imprecise and complete analysis, as well as incorrect judgment. SUMMARY OF THE INVENTION Therefore, the inventors have invented an artificial intelligence analysis identification prediction method based on various shortcomings in the current market to identify the shortcomings of the lack of self-correcting identification and analysis rules for soft and hard bodies. The main purpose of the present invention is to provide an artificial intelligence analysis and identification prediction method to establish a set of intelligent identification, analysis and prediction criteria, and to identify and predict a target, and give a prediction result. Then the result of the ratio and the predicted result, the corrected, the accuracy of the measurement. - The identification rule is improved to achieve the above purpose, and the present invention is to make the aforementioned person JL know the body of the eight batches to be installed in a hard body type. An electronic computer hard next step: the worker wisdom line to execute its money, and contains the wheeled data of the multiple pens, wherein each piece of data contains the overflow value and - _; 聿贝^ has a plurality of special ones The built-in machine learning algorithm learns the input data; according to the learning result of the machine learning algorithm, the model of the prediction target is established. ^^^ Knife analysis and identification::: The group of factors that predict the variation of the target result And the impact: I: the weight of the impact weight; I input the text of the forecast target, 1 eigenvalue; - the shell contains a plurality of ~ use the pattern and rules to identify and analyze the forecast Data and push 4 in the category of the target; round the target in the de facto category of light to the class W and compare it to the category of the fact and the previous category, if the two are the same, then the end Method, if the comparison is different, then the following multiple learning steps are performed: It is expected that a new capital will be generated based on the information of the forecast target and the de facto category.
5 J288332 以機器學習演算法學習該新的資料； 根據機器學習演算法的演算結果增減分別可影響各類 別的各因素，改變各因素的權重，且校正該預測及分析預 測目標之模式與規則。 藉由上述技術手段，機器學習演算法可在本方法中建 立一模式及規則以辨識及分析該預測目標，並藉由錯誤的 預測結果及事實上的結果來修正該模式及規則，藉此提高 預測的精確度。 • 【實施方式】 請參照第一、二及三圖，本發明人工智慧分析辨識預 測方法，係以軟體型態安裝在一電子計算機硬體内，以成 為一人工智慧系統來執行其功能，且包含有以下步驟： 輸入複數筆負料’其中各筆資料含有複數個特徵值及 一個類別（3 0 1 );例如，輸入資料1資料2 ···到 資料η等η筆資料於系統中，其中各筆資料可為一筆基因 序列’且所包含的特徵值係為擷取自該基因序列的第丨基 _ 因—第m基因等m筆基因之值，而該類別為個體細胞變異 之結果，例如該個體細胞變異之結果係共有二類別，分別 為「患有肺癌之細胞與正常細胞」或者有三類別，分別為 「患有AML型態白血症之細胞與、患有ALL型態白血症之 細胞與患有MLL型態白血症之細胞」，或者有更多類別。 以内建的機器學習演算法（Machine learning Algorithm)學習所輸入資料（3〇2);其中機器學習 演算法在電腦程式語言的領域中具有多種態樣，例如 (§： 6 -1288332 B〇〇tStraPping-bo〇sting Algorithm，且依實際應用目的 及所搭配編塔哭f . w σ ( comp i ier，如 c 語言及 Visual Basic 等等）之不同而在撰寫成軟體時有所不同，故不另贅述。 根據機器學習演算法的學習結果，建立一可分析及辨 識預測目^之模式與規則（patterns and rules )( 3 〇 3 )，並自該模式與規則中，擷取可影響預測目標結果 隻異之因素群及影響能力大小，即抽取自模式與規則中的 複數個特徵點（因素群）以及各特徵點（因素群）所分別 擁有的衫響權重值（影響能力大小），且經由該模式與規 則自與各類別相對應的複數個特徵點之中，選出其中至 )-個可影響該類別的特徵點，以作為影響該類別的至少 因素（303a)，以及各因素所分別擁有的影響權重 值（303b);例如，前述的㈤筆基因值中，其中第工、 3、5 ' 7及9基因等五筆基因點係為具有影響罹患肺癌 的因素’而該五筆的權重值可分別為2〇%、2〇%、2 〇%'10%及30%’而模式與規則為前述機器學習演 算法自動產生之多元多次程式。 、 輸入-預測目標之資料，其中該資料包含複數個特徵 (3 0 4 )’例如’輸入一筆某人的基因資料，該資料 含有第1 —m基因之值。 以該模式與規則來辨識與分析該預測目標資料（3 〇 5) ’並根據分析結果來推得該目標之類別；即將目標資 料中的各項基因值代入前述的數學運算方法中，根據運算 結果將該筆資料歸類於「患有肺癌」，或者「沒有肺癌」 -1288332 的類別。 6 ·輸入該目標於事實上之類別，並 與前述推得之類別（3 0 6 )，即得知事實上類別 實患有肺癌，並與系統所推得的「患有肺癌」二：= 有肺癌」之類別相比對’若兩者比對正確則結束本方^ 07)，兩者若比對不同，則進行下列學習步驟： 將該預測目標的資料及事實上的 ( 3 0 8 )。 1聿新貝枓 以機器學習演算法學習該新的資料（3 〇 9 ) :據機器學習演算法的演算結果增響 又變各因素的權重值（310 b )，且校正該預測及分析 * ^ ^ Π c ) . ^目 式與規則（3 i 〇 )，例如，原先歸納出的第1、3、5、7及9基因 對罹患癌症的類別具有影響及9基因 ^ ^ ^ 1 一、、&過修正後，重新推得且 有衫響該類別的基因新增了一 - 篦弓美因避舌姓* 個第1 1基因，且修正後的 第5基因權重值降為1()%， 〇%。 而第11基因之權重值為25 J288332 Learning the new data with a machine learning algorithm; according to the calculation results of the machine learning algorithm, the factors affecting each category, changing the weight of each factor, and correcting the prediction and analysis of the prediction target model and rules . By the above technical means, the machine learning algorithm can establish a pattern and rules in the method to identify and analyze the predicted target, and correct the pattern and rules by erroneous prediction results and de facto results, thereby improving The accuracy of the forecast. • [Embodiment] Please refer to the first, second and third figures. The artificial intelligence analysis identification prediction method of the present invention is installed in an electronic computer hardware in a software type to be an artificial intelligence system to perform its function, and The method includes the following steps: inputting a plurality of negative materials, wherein each piece of data has a plurality of characteristic values and a category (3 0 1 ); for example, inputting data 1 data 2 ··· to data η and other data in the system, Each of the data may be a gene sequence 'and the characteristic value included is the value of the m-th gene derived from the 丨 _ _ m gene of the gene sequence, and the category is the result of individual cell variability For example, the results of the individual cell variability are divided into two categories, namely, "cells with lung cancer and normal cells" or three categories, respectively, "cells with AML type white blood, and with ALL type of white blood Cells and cells with MLL type white bloodemia, or more categories. The built-in machine learning algorithm (3) is used to learn the input data (3〇2); the machine learning algorithm has many aspects in the field of computer programming language, for example (§: 6 -1288332 B〇〇tStraPping -bo〇sting Algorithm, and depending on the actual application purpose and the matching tower crying f. w σ (com i ier, such as c language and Visual Basic, etc.), it is different when writing software, so there is no other Describing. According to the learning result of the machine learning algorithm, a pattern and rules (3 〇 3 ) can be constructed and analyzed, and from the pattern and rules, the prediction results can be predicted. The difference between the factor group and the influencing ability, that is, the number of feature points (factor groups) extracted from the pattern and the rule, and the weight of the shirt (the magnitude of the influence) possessed by each feature point (factor group), and The pattern and the rule are selected from among a plurality of feature points corresponding to each category, and the feature points that affect the category are selected as at least factors affecting the category. 303a), and the influence weight value (303b) of each factor; for example, among the above (5) gene values, the five genes of the work, 3, 5 '7 and 9 genes are affected by lung cancer. The factor 'and the weight value of the five strokes can be 2〇%, 2〇%, 2〇% '10% and 30%, respectively, and the pattern and rules are the multiple multiple programs automatically generated by the aforementioned machine learning algorithm. And input-predicted target data, wherein the data includes a plurality of features (3 0 4 )', for example, 'Entering a person's genetic data, the data containing the value of the first-m gene. The model and the rule are used to identify and analyze the predicted target data (3 〇 5) ' and the category of the target is derived according to the analysis result; that is, each gene value in the target data is substituted into the aforementioned mathematical operation method, according to the operation As a result, the data was classified as "with lung cancer" or "no lung cancer" -1288332. 6 · Enter the target in the de facto category and the category (3 0 6 ) that was derived from the above, that is, the fact that the category actually has lung cancer, and the system has the "lack of lung cancer" two: If there is a category of lung cancer, the comparison is made to 'If the two are correct, the party ends ^ 07.) If the two are different, the following learning steps are performed: The data of the prediction target and the factual ( 3 0 8 ). 1聿新贝枓 uses the machine learning algorithm to learn the new data (3 〇 9 ): According to the calculation result of the machine learning algorithm, the increase and change of the weight value of each factor (310 b ), and correct the prediction and analysis * ^ ^ Π c ) . ^Objectives and rules (3 i 〇), for example, the originally derived genes 1, 3, 5, 7 and 9 have an effect on the type of cancer and 9 genes ^ ^ ^ 1 I, &After the correction, the gene that has been re-twisted and has a shirt in this category has added a new one - 篦 美 美 美 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避 避, 〇%. The weight of the 11th gene is 2
除了前述運用本發明I 讲办外令叮 來預匐基因與疾病之關係的基因 研九外，亦可運用本發明在人臉 聲紋之辨識，以進行門杯祝构膘m 严辨喷β ATM晶片卡及信用卡等身 伤辨識，或者是運用本發 用鼠險蟬枋♦ "於財務資訊的辨識，以進行信 用風險汗核之預測及金融財務預測等等。 與像=辨識為例’可輪入複數筆人臉影像資料， 為靜態影像、三維動態影像或是四維立趙動 8 1288332 μ〜像，該等特徵值係為眼、耳、鼻、口等擷取自該影像 的局部影像特徵值，且具有—該張人臉的身分類別，最後 輸入預測目標的人臉影像至系統内可得到一預測的身分。 以聲紋辨識為例，可輸入複數筆聲譜資料，各筆聲譜 具有數個局部聲譜特徵值，且具有一該筆聲譜的身分類 別，最後輸入預測目標的聲譜至系統内可得到一預測的身 分。 以^用財務之預測為例，所輸入該系統的複數筆資料 中’各筆資料分別為一個體基本資料以及過去信用狀況資 料以及信用風險值，該等特徵㈣為複數個#|取自個體資 料的不同基本資料以及過去信用狀況資料，該類別為信用 風險值，可為多種類類別資料。 藉由上述技術手段，機器學習演算法在本方法中所建 立的模式及規則係可辨識及分析該預測目#，並藉由錯誤 的預測結果及事實上的結果來修正該模式及規則，藉此提 高預測的精確度。 【圖式簡單說明】 第-圖係為本發明人工智慧分析辨識預測方法以軟體 塑態安裝於一電子機算機硬體之示意圖。 第二圖係為本發明所輸人到“之單筆資料的示意 圖。 第三圖人工智慧分析辨識預測方法之步驟示意圖。 【主要元件符號說明】 (3 0 1 )輸入複數筆資料於系統 9 J288332 (3 Ο 2 )已機器學習演算法學習所輸入之資料 (3 0 3 )建立一辨識及分析之模式及規則 (3 0 3 a )建立各類別中影響該類別的因素 (3 0 3 )建立一辨識及分析之模式與規則 (3 0 3 b )建立各類別中可影響該類別之因素的權 重值 (3 0 4 )對系統輸入預測目標之資料 (3 0 5 )以模式與規則來辨識與分析該預測目標並 _ 推得該目標所屬類別 (3 0 6 )比對所推得類別與事實類別 (3 0 7 )結束 ’ (3 0 8 )將預測目標資料及事實類別作為一筆新資 . 料 (3 0 9 )以機器學習演算法學習該筆新資料 (3 1 0 a )增減各類別相對應之因素 (3 1 0 b )改變各類別相對應之因素的權重值 # ( 3 1 0 c )校正模式與規則In addition to the above-mentioned application of the present invention, I can use the invention to predict the relationship between genes and diseases, and can also use the invention to identify the voiceprint of the human face, so as to carry out the door cup and the structure of the door. Identification of injuries such as ATM chip cards and credit cards, or the use of the identification of the company's financial risks ♦ " in the identification of financial information for credit risk Khan nuclear forecast and financial and financial forecasts. With the image = identification as an example, you can enter multiple face image data, which are still images, 3D motion pictures or 4D Vientiane 8 1288332 μ~ images, such as eyes, ears, nose, mouth, etc. The local image feature value of the image is taken, and has the identity type of the face, and finally the face image of the predicted target is input into the system to obtain a predicted identity. Taking voiceprint recognition as an example, a plurality of sound spectrum data can be input, each sound spectrum has several local sound spectrum feature values, and has an identity class of the pen sound spectrum, and finally the sound spectrum of the predicted target is input into the system. Get a predictive identity. Taking the financial forecast as an example, in the multiple data input into the system, 'each data is a basic data and past credit status data and credit risk value, and these features (4) are plurals #| taken from the individual Different basic data of the data and past credit status data, this category is the credit risk value, which can be a variety of categories of information. Through the above technical means, the pattern and rule established by the machine learning algorithm in the method can identify and analyze the prediction target #, and correct the pattern and rules by erroneous prediction results and de facto results. This improves the accuracy of the prediction. [Simple diagram of the diagram] The first diagram is a schematic diagram of the artificial intelligence analysis identification prediction method of the invention installed in a soft plastic state on an electronic computer hardware. The second picture is a schematic diagram of the single data input by the invention. The third figure is a schematic diagram of the steps of the artificial intelligence analysis identification prediction method. [Key component symbol description] (3 0 1 ) Input complex data in system 9 J288332 (3 Ο 2 ) The data entered by the machine learning algorithm learning (3 0 3 ) establishes a pattern and rules for identification and analysis (3 0 3 a ) to establish the factors affecting the category in each category (3 0 3 ) Establish a pattern and rules for identification and analysis (3 0 3 b ) to establish the weight value of each category that can affect the category (3 0 4 ). The data (3 0 5 ) of the system input prediction target is based on patterns and rules. Identify and analyze the predicted target and _ push the target category (3 0 6 ) to compare the predicted category with the fact category (3 0 7 ) to end ' (3 0 8 ) to predict the target data and fact categories as a new (3 0 9 ) Learning the new data by machine learning algorithm (3 1 0 a ) increasing or decreasing the corresponding factors of each category (3 1 0 b ) changing the weight value of the corresponding factors of each category# ( 3 1 0 c ) Calibration mode and rules