convolutional networks, won 11% of games against Pachi23 and 12% against a slightly weaker program, Fuego24. Reinforcement learning of value networks The final stage of the training pipeline focuses on position evaluation, (s, a) of the search tree stores an action value Q(s, a), visit count N(s, a), and prior probability P(s, a). The tree is traversed by simulation (that is, descending the tree in complete games without backup), starting from the root state. At each time step t of each simulation, an action at is selected from state st Figure 3 | Monte Carlo tree search in AlphaGo. a, Each simulation traverses the tree by selecting the edge with maximum action value Q, plus a bonus u(P) that depends on a stored prior probability P for that edge. b, The leaf node may be expanded; the new node is processed once by the policy network pσ and the output probabilities are stored as prior probabilities P for each action. c, At the end of a simulation, the leaf node is evaluated in two ways: using the value network vθ ; and by running a rollout to the end of the game with the fast rollout policy pπ , then computing the winner with function r. d, Action values Q are updated to track the mean value of all evaluations r(·) and vθ (·) in the subtree below that action. Selection a b c d Expansion Evaluation Backup p S p V Q + u(P) Q + u(P) Q + u(P) Q + u(P) P P P P Q Q Q Q Q r r r r P max max P Q T Q T Q T Q T Q T Q T
combine the content of a photograph with the style of several well-known artworks. The images were created by ﬁnding an image that simultaneously matches the content representation of the photograph and the style representation of the artwork (see Methods). The original photograph depicting the Neckarfront in T¨ ubingen, Germany, is shown in A (Photo:
r net with this 3-layer bottleneck block, resulting in yer ResNet (Table 1). We use option B for increasing ions. This model has 3.8 billion FLOPs. -layer and 152-layer ResNets: We construct 101- nd 152-layer ResNets by using more 3-layer blocks 1). Remarkably, although the depth is signiﬁcantly ed, the 152-layer ResNet (11.3 billion FLOPs) still wer complexity than VGG-16/19 nets (15.3/19.6 bil- OPs). 50/101/152-layer ResNets are more accurate than layer ones by considerable margins (Table 3 and 4). not observe the degradation problem and thus en- niﬁcant accuracy gains from considerably increased The beneﬁts of depth are witnessed for all evaluation (Table 3 and 4). arisons with State-of-the-art Methods. In Table 4 mpare with the previous best single-model results. seline 34-layer ResNets have achieved very compet- curacy. Our 152-layer ResNet has a single-model alidation error of 4.49%. This single-model result orms all previous ensemble results (Table 5). We e six models of different depth to form an ensemble method error (%) Maxout  9.38 NIN  8.81 DSN  8.22 # layers # params FitNet  19 2.5M 8.39 Highway [42, 43] 19 2.3M 7.54 (7.72±0.16) Highway [42, 43] 32 1.25M 8.80 ResNet 20 0.27M 8.75 ResNet 32 0.46M 7.51 ResNet 44 0.66M 7.17 ResNet 56 0.85M 6.97 ResNet 110 1.7M 6.43 (6.61±0.16) ResNet 1202 19.4M 7.93 Table 6. Classiﬁcation error on the CIFAR-10 test set. All meth- ods are with data augmentation. For ResNet-110, we run it 5 times and show “best (mean±std)” as in . so our residual models have exactly the same depth, width, and number of parameters as the plain counterparts. We use a weight decay of 0.0001 and momentum of 0.9, ग़ॴɿhttps://www.robots.ox.ac.uk/~vgg/rg/papers/deepres.pdf 110Ͱ170ສύϥϝλʂ
(http://www.image-net.org/) Ͱެ։͍ͯ͠Δը૾ 1400ສຕ ɹ→ ͜ͷେྔσʔλͷֶशͷͨΊʹ GPU ͕ඞਢ ग़ॴɿhttp://static.googleusercontent.com/media/research.google.com/ja//archive/unsupervised_icml2012.pdf Building high-level features using large-scale unsupervised the cortex. They also demonstrate that convolutional DBNs (Lee et al., 2009), trained on aligned images of faces, can learn a face detector. This result is inter- esting, but unfortunately requires a certain degree of supervision during dataset construction: their training images (i.e., Caltech 101 images) are aligned, homoge- neous and belong to one selected category. Figure 1. The architecture and parameters in one layer of our network. The overall network replicates this structure three times. For simplicity, the images are in 1D. logical and computation Lyu & Simoncelli, 2008; As mentioned above, cen of local connectivity bet ments, the ﬁrst sublayer pixels and the second su lapping neighborhoods o The neurons in the ﬁrst su input channels (or maps second sublayer connect (or map).3 While the ﬁrs responses, the pooling lay the sum of the squares o is known as L2 pooling. Our style of stacking ules, switching betwe ance layers, is remini HMAX (Fukushima & M 1998; Riesenhuber & Po been argued to be an a brain (DiCarlo et al., 201 Although we use local not convolutional: the across diﬀerent location a stark diﬀerence betw vious work (LeCun et a Building high-level features using large-scale unsupervised learning and minimum activation values, then picked 20 equally spaced thresholds in between. The reported accuracy is the best classiﬁcation accuracy among 20 thresholds. 4.3. Recognition Surprisingly, the best neuron in the network performs very well in recognizing faces, despite the fact that no supervisory signals were given during training. The best neuron in the network achieves 81.7% accuracy in detecting faces. There are 13,026 faces in the test set, so guessing all negative only achieves 64.8%. The best neuron in a one-layered network only achieves 71% ac- curacy while best linear ﬁlter, selected among 100,000 ﬁlters sampled randomly from the training set, only achieves 74%. To understand their contribution, we removed the lo- cal contrast normalization sublayers and trained the network again. Results show that the accuracy of best neuron drops to 78.5%. This agrees with pre- vious study showing the importance of local contrast normalization (Jarrett et al., 2009). We visualize histograms of activation values for face images and random images in Figure 2. It can be seen, even with exclusively unlabeled data, the neuron learns to diﬀerentiate between faces and random distractors. Speciﬁcally, when we give a face as an input image, the neuron tends to output value larger than the threshold, 0. In contrast, if we give a random image as an input image, the neuron tends to output value less than 0. Figure 2. Histograms of faces (red) vs. no faces (blue). The test set is subsampled such that the ratio between faces and no faces is one. tested neuron, by solving: x∗ = arg min x f(x; W, H), subject to ||x||2 = 1. Here, f(x; W, H) is the output of the tested neuron given learned parameters W, H and input x. In our experiments, this constraint optimization problem is solved by projected gradient descent with line search. These visualization methods have complementary strengths and weaknesses. For instance, visualizing the most responsive stimuli may suﬀer from ﬁtting to noise. On the other hand, the numerical optimization approach can be susceptible to local minima. Results, shown in Figure 3, conﬁrm that the tested neuron in- deed learns the concept of faces. Figure 3. Top: Top 48 stimuli of the best neuron from the test set. Bottom: The optimal stimulus according to nu- merical constraint optimization. 4.5. Invariance properties GoogleͷೣͷจͰ 1000ສຕͷ Youtube ը૾Λ༻
ɾ ɾ ɾ ɾ ɾ ɾ ɾ ɾ Dropout, on the other hand, prevents overﬁtting, even in this case. It does not even need early stopping. Goodfellow et al. (2013) showed that results can be further improved to 0.94% by replacing ReLU units with maxout units. All dropout nets use p = 0.5 for hidden units and p = 0.8 for input units. More experimental details can be found in Appendix B.1. Dropout nets pretrained with stacks of RBMs and Deep Boltzmann Machines also give improvements as shown in Table 2. DBM—pretrained dropout nets achieve a test error of 0.79% which is the best performance ever reported for the permutation invariant setting. We note that it possible to obtain better results by using 2-D spatial information and augmenting the training set with distorted versions of images from the standard training set. We demonstrate the e↵ectiveness of dropout in that setting on more interesting data sets. With dropout Without dropout @ R @ @ R In order to test the robustness of dropout, classiﬁcation experiments were done with networks of many di↵erent ar- chitectures keeping all hyperparameters, in- cluding p, ﬁxed. Figure 4 shows the test error rates obtained for these di↵erent ar- chitectures as training progresses. The same architectures trained with and with- out dropout have drastically di↵erent test errors as seen as by the two separate clus- ters of trajectories. Dropout gives a huge improvement across all architectures, with- out using hyperparameters that were tuned speciﬁcally for each architecture.
classes auto horse dog truck New test image from unknown class cat Training images Figure 1: Overview of our cross-modal zero-shot model. We ﬁrst map each new testing image into a lower dimensional semantic word vector space. Then, we determine whether it is on the manifold of seen images. If the image is ‘novel’, meaning not on the manifold, we classify it with the help of
CNN ͱͷ߹ͤͳͲΜʹݚڀ͞Ε͍ͯΔ ग़ॴɿhttps://arxiv.org/pdf/1411.4555.pdf ɹɹ https://arxiv.org/pdf/1604.04573.pdf al Image Caption Generator Samy Bengio Google firstname.lastname@example.org Dumitru Erhan Google email@example.com A group of people shopping at an outdoor market. ! There are many vegetables at the fruit stand. Vision! Deep CNN Language ! Generating! RNN Figure 1. NIC, our model, is based end-to-end on a neural net- work consisting of a vision CNN followed by a language gener- ating RNN. It generates complete sentences in natural language from an input image, as shown on the example above. CNN-RNN: A Uniﬁed Framework for Multi-label Image Classiﬁcation Jiang Wang1 Yi Yang1 Junhua Mao2 Zhiheng Huang3∗ Chang Huang4∗ Wei Xu1 Baidu Research 2University of California at Los Angles 3Facebook Speech 4 Horizon Robotics Abstract While deep convolutional neural networks (CNNs) have own a great success in single-label image classiﬁcation, s important to note that real world images generally con- n multiple labels, which could correspond to different jects, scenes, actions and attributes in an image. Tradi- nal approaches to multi-label image classiﬁcation learn dependent classiﬁers for each category and employ rank- g or thresholding on the classiﬁcation results. These tech- ques, although working well, fail to explicitly exploit the Airplane Great Pyrenees Archery Sky, Grass, Runway Dog, Person, Room Person, Hat, Nike Figure 1. We show three images randomly selected from ImageNet 2012 classiﬁcation dataset. The second row shows their corre- sponding label annotations. For each image, there is only one la- bel (i.e. Airplane, Great Pyrenees, Archery) annotated in the Im-