This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
binary_network [2018/04/20 15:04]
binary_network [2018/11/04 13:10]
Line 57: Line 57:
 https://​github.com/​cair/​TsetlinMachine The Tsetlin Machine - A Game Theoretic Bandit Driven Approach to Optimal Pattern Recognition with Propositional Logic https://​github.com/​cair/​TsetlinMachine The Tsetlin Machine - A Game Theoretic Bandit Driven Approach to Optimal Pattern Recognition with Propositional Logic
 +https://​arxiv.org/​pdf/​1805.04908.pdf On the Practical Computational Power of Finite Precision RNNs
 +for Language Recognition
 +In particular, we show that the
 +LSTM and the Elman-RNN with ReLU
 +activation are strictly stronger than the
 +RNN with a squashing activation and the
 +GRU. This is achieved because LSTMs
 +and ReLU-RNNs can easily implement
 +counting behavior. We show empirically
 +that the LSTM does indeed learn to
 +effectively use the counting mechanism.
 +https://​arxiv.org/​abs/​1806.07550v1 Binary Ensemble Neural Network: More Bits per Network or More Networks per Bit?
 +While ensemble techniques have been broadly believed to be only marginally helpful for strong classifiers such as deep neural networks, our analyses and experiments show that they are naturally a perfect fit to boost BNNs. We find that our BENN, which is faster and much more robust than state-of-the-art binary networks, can even surpass the accuracy of the full-precision floating number network with the same architecture.
 +https://​arxiv.org/​abs/​1809.03368v1 Probabilistic Binary Neural Networks
 +Low bit-width weights and activations are an effective way of combating the increasing need for both memory and compute power of Deep Neural Networks. In this work, we present a probabilistic training method for Neural Network with both binary weights and activations,​ called BLRNet. By embracing stochasticity during training, we circumvent the need to approximate the gradient of non-differentiable functions such as sign(), while still obtaining a fully Binary Neural Network at test time. Moreover, it allows for anytime ensemble predictions for improved performance and uncertainty estimates by sampling from the weight distribution. Since all operations in a layer of the BLRNet operate on random variables, we introduce stochastic versions of Batch Normalization and max pooling, which transfer well to a deterministic network at test time. We evaluate the BLRNet on multiple standardized benchmarks.
 +https://​arxiv.org/​abs/​1809.04547 . Using the Tsetlin Machine to Learn Human-Interpretable Rules for High-Accuracy Text Categorization with Medical Applications
 +In all brevity, we represent the terms of a text as propositional variables. From these, we capture categories using simple propositional formulae, such as: if "​rash"​ and "​reaction"​ and "​penicillin"​ then Allergy. The Tsetlin Machine learns these formulae from a labelled text, utilizing conjunctive clauses to represent the particular facets of each category. Indeed, even the absence of terms (negated features) can be used for categorization purposes. Our empirical results are quite conclusive. The Tsetlin Machine either performs on par with or outperforms all of the evaluated methods on both the 20 Newsgroups and IMDb datasets, as well as on a non-public clinical dataset. On average, the Tsetlin Machine delivers the best recall and precision scores across the datasets. The GPU implementation of the Tsetlin Machine is further 8 times faster than the GPU implementation of the neural network.
 +https://​arxiv.org/​pdf/​1809.09244.pdf No Multiplication?​ No Floating Point? No Problem!
 +Training Networks for Efficient Inference
 +we train deep networks that emit only a
 +predefined, static number of discretized values. Despite reducing the number of
 +values that can be emitted from 2
 +32 to only 32, there is little to no degradation in
 +network performance across a variety of tasks. Compared to existing approaches
 +for discretization,​ our approach is both conceptually and programmatically simple
 +and has no stochastic component. Second, we provide a method to constrain the
 +network’s weights to a small number of unique values (typically 100-1000) by
 +employing a periodic adaptive clustering step during training.
 +https://​arxiv.org/​abs/​1810.03538v1 Combinatorial Attacks on Binarized Neural Networks
 +The discrete, non-differentiable nature of BNNs, which distinguishes them from their full-precision counterparts,​ poses a challenge to gradient-based attacks. In this work, we study the problem of attacking a BNN through the lens of combinatorial and integer optimization. ​
 +https://​arxiv.org/​pdf/​1803.07125v2.pdf Local Binary Pattern Networks