LSTM 智能导向的电子信息实践教改探索

Exploration of LSTM intelligence-oriented electronic information practice education reform

针对传统电子信息专业教学中存在的问题,提出融合AI技术的针对性LSTM智能导向实践教学模型,构建了因材施教的人才培养体系。结合AI技术构建融合平台,包括结合深度学习技术在电子信息基础理论与实践课程中有针对性地推荐课程;借鉴强化学习思想和组内竞争思想助力学生竞赛创新;采用LSTM对学生个体进行建模,用以准确评估学生的自身能力,以便教师对他们进行因材施教。该教改方案实施以来,实验班相比对照班在获奖门类和数量上都有显著提高,验证了所提教改方案的有效性。

[Objective]The combination of digital education and iterative upgrading of traditional education models has achieved good experimental results.Using digital empowerment to reform teaching models and intelligent overall management,diversified teaching models can be promoted and the efficiency of education operation and management can be fundamentally improved.In response to problems existing in the course design and teaching process of electronic information majors,this article builds an integrated teaching platform and proposes a quantitative student algorithm combined with long short-term memory(LSTM)technology to assist teachers in providing targeted training for students.[Methods]By modeling individual student models and using LSTM training,a student development model is constructed with individual students as the subject and parameters such as the assessment of students'motivation in class,records of correctness of practice problems,and awards for participation in competitions.Drawing on artificial intelligence conceptions,the intelligent prediction data process is used for practical education reform,and one-hot coding is used to encode individual student data.In addition,for the characteristics of individual student training data,an attention mechanism is introduced to the LSTM model,and a neural network is added to each state output of the original LSTM model.It makes the LSTM model selectively rely on all previous data records of the students instead of only relying on a single data input from the previous step when predicting the output of the students'data,which can make the prediction of the output results more accurate.In this paper,the LSTM network is combined with attention mechanisms to input all time steps of the LSTM hidden states and then input sequences into an attention model to compute the attention weights for each time step.Then,the computed attention weights are added to the input sequences to obtain a weighted input vector,which is fed into the LSTM network to perform the next prediction step.The model predicts individual student needs by learning a large amount of student data;ultimately,students can be taught according to their needs based on more accurate localization of predictions.[Results]By collecting student learning data and training models,the learning path and competition potential of student subject exercises can be accurately predicted.The experimental results compared with a contrast class show the following.1)After three years of practice,an experimental class and a contrast one were set up to compare the results.After implementing the new teaching reform based on the above materials,the experimental class showed 14.6%,15.4%,and 180%improvement in the daily performance,final performance,and the number of competition results compared with those of the contrast class,respectively.2)The number of admitted graduate students and their interest in scientific research significantly increased,and the anticipated employment is widely welcomed by enterprises.[Conclusions]By integrating teaching reform models and focusing on evolving LSTM models to quantitatively predict student development data,an electronic information practical teaching reform plan is proposed.The use of LSTM to model the learning outcomes of students allows the accurate evaluation of their own abilities and provides teachers with objective teaching plans.Therefore,the improved method in this article achieved the expected goals to help teachers reflect on the effectiveness of educational reform,enhance their ability to understand and specialize in research,and thus significantly solve educational reform problems.