來源：專知

概要：第15屆中國機器學習及其應用研討會今天11月4日在北京交通大學舉行，海內外從事機器學習及相關領域研究的10餘位專家與會進行學術交流。

第15屆中國機器學習及其應用研討會今天11月4日在北京交通大學舉行，海內外從事機器學習及相關領域研究的10餘位專家與會進行學術交流，包括特邀報告、頂會論文交流、以及Top Conference Review等部分。

1. 深度森林初探

這是由機器學習西瓜書作者、南京大學周志華老師講述的關於他最新整合學習研究成果-深度森林，一種對深度神經網路可替代性方法。

圖示：級聯森林結構的圖示。級聯的每個級別包括兩個隨機森林（藍色字型標出）和兩個完全隨機樹木森林（黑色）。假設有三個類要預測; 因此，每個森林將輸出三維類向量，然後將其連線以重新表示原始輸入。

gcForest的整體架構

gcForest在影象分類、人臉識別、音樂分類、情感分類等特定資料集上都取得了非常不錯的分類效果，是非深度神經網路表現最好的方法。gcForest只是深度森林一個開始。有很多可探索的可能性和應用場景。

2. Latent tree analysis

香港科技大學張連文教授的報告。

Latent tree analysis seeks to model the correlations among a set of random variables using a tree of latent variables. It was proposed as an improvement to latent class analysis — a method widely used in social sciences and medicine to identify homogeneous subgroups in a population. It provides new and fruitful perspectives on a number of machine learning areas, including cluster analysis, topic detection, and deep probabilistic modeling. In this talk, I will give an overview of the research on latent tree analysis and various ways it is used in practice.

3. Graph Refinement

浙江大學張振躍教授的報告。

資料聚類方法的有效性非常受制於差異性或相似性圖矩陣內涵的類屬性特點。由於受多種因素的影響，圖矩陣或高維資料本身的類屬性通常比較模糊，即便是由區域性鄰域點構成的圖矩陣也通常如此。在多源異尺度資料聚類中，圖矩陣的類屬性模糊性或矛盾性更為明顯。在本報告中，我們將從三個角度考慮如何修正給定的圖矩陣，提升圖矩陣的類屬性：（1）從多源資料的視角扭曲及圖矩陣形模擬，恢復固有的一致性圖矩陣；（2）從多源資料的稀疏鄰域表達，構建一致化稀疏圖矩陣；（3）從單源圖矩陣的稀疏低秩逼近，修正圖矩陣 。我們將從理論基礎、模型建立、演算法設計和數值檢驗等方面說明上述圖修正方法的合理及其有效性。

4. Low-dimensional Structures and Deep Models for High-dimensional (Visual) Data

加州大學伯克利分校馬毅教授的報告。

We  discuss a class of models and techniques that can effectively model and extract rich low-dimensional structures in high-dimensional data such as images and videos, despite nonlinear transformation, gross corruption, or severely compressed measurements. This work leverages recent advancements in convex optimization from Compressive Sensing for recovering low-rank or sparse signals that provide both strong theoretical guarantees and efficient and scalable algorithms for solving such high-dimensional combinatorial problems. We illustrate how these new mathematical models and tools could bring disruptive changes to solutions to many challenging tasks in computer vision, image processing, and pattern recognition. We will also illustrate some emerging applications of these tools to other data types such as 3D range data, web documents, image tags, bioinformatics data, audio/music analysis, etc. Throughout the talk, we will discuss strong connections of algorithms from Compressive Sensing with other popular data-driven models such as Deep Neural Networks, providing some new perspectives to understand Deep Learning. 

5. 回覆神經網路學習

四川大學張蕾教授的報告。

隨著大資料時代的到來及深度神經網路的興起，神經網路在影象理解、語音識別、自然語言處理等領域取得了令人矚目的成功。回覆神經網路作為神經網路的一種主要用於處理時序資料，廣泛用於機器翻譯、影象理解、情感分析、語音翻譯等時序任務中。這一講座將系統地對回覆神經網路進行回顧，並針對其兩個學習演算法Back Propagation Through Time (BPTT) 和Real Time Recurrent Learning (RTRL) 進行介紹，並基於此對回覆神經網路訓練中存在的問題進行了“進一步的思考”。具體包括：（1）生物神經網路與人工神經網路；（2）回覆神經網路的學習演算法BPTT和RTRL；（3）回覆神經網路訓練過程中存在的“梯度消失”問題及相應的解決方法，基於此簡要地介紹新的回覆神經網路模型，如：Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) 及 Recurrent Highway Network (RHN)等。

6. Towards Understanding Deep Learning: Two Theories of Stochastic Gradient Langevin Dynamics

北京大學王立威教授的報告。

Deep learning has achieved great success in many applications. However, deep learning is a mystery from a learning theory point of view. In all typical deep learning tasks, the number of free parameters of the networks is at least an order of magnitude larger than the number of training data. This rules out the possibility of using any model complexity-based learning theory (VC dimension, Rademacher complexity etc.) to explain the good generalization ability of deep learning. Indeed, the best paper of ICLR 2017 “Understanding Deep Learning Requires Rethinking Generalization” conducted a series of carefully designed experiments and concluded that all previously well-known learning theories fail to explain the phenomenon of deep learning.

7.  大規模分類任務的結構化學習策略

胡清華 教授 天津大學

隨著資料規模的不斷擴大，分類學習演算法面臨的任務也越來越複雜，分類學習的類別數從幾個增長到幾百個，甚至幾萬個。此時，不同的類別標籤之間可能會形成複雜的結構關係。充分利用這種結構資訊可顯著提升分類效能和決策的可靠性。本報告將討論結構化學習任務的特點、評價指標、特徵評價和分類模型構造演算法。

8. Active Learning: Query Less for More

黃聖君 副教授 南京航空航天大學

 In supervised learning, a large training set of labeled examples is usually required to train an effective model. However, in many real applications, there are plentiful unlabeled data but limited labeled data, and the acquisition of labels is costly. Active learning reduces the labeling cost by iteratively selecting the most valuable data to query their labels from the annotator. This talk will summarize some important issues in active learning, including the designing of selection criterion and query type, querying from imperfect annotators and fast selection from large scale unlabeled data. Our recent efforts towards solving these issues will be reported.