Brendan McCane

Brendan McCane

The Legg--Hutter universal intelligence measure provides a rigorous scalar assessment of general intelligence as expected reward across all computable environments, weighted by simplicity. However, the measure presupposes an externally specified reward function, raising the question of whether the reward primitive is inherently arbitrary or whether a canonical choice exists. We propose a new measure, called intervention complexity, that has five natural properties: environment-derivedness, universality, minimality, sensitivity, and achievement preference. Given a resource function rho encoding an inductive bias (such as program length, execution time, or energy), rho-intervention complexity is a universal reward. The result yields a family of canonical rewards indexed by resource bias, providing a principled completion of the Legg--Hutter framework that does not require external normative input. We further propose a two-dimensional characterisation of intelligence: agent competence (how well the agent performs relative to the oracle optimum) and learning efficiency (how quickly this competence improves with experience). A separation theorem establishes that the choice of resource bias determines the computability of the resulting measure: action-count IC is computable in polynomial time, while program-length IC without oracle access is uncomputable, with the gap between oracle and bare IC precisely quantifying the information-theoretic content of learning. We discuss implications for superintelligence and for pre-training universal agents.

Lech Szymanski, Brendan McCane, Craig Atkinson

We propose a complexity measure of a neural network mapping function based on the diversity of the set of tangent spaces from different inputs. Treating each tangent space as a linear PAC concept we use an entropy-based measure of the bundle of concepts in order to estimate the conceptual capacity of the network. The theoretical maximal capacity of a ReLU network is equivalent to the number of its neurons. In practice however, due to correlations between neuron activities within the network, the actual capacity can be remarkably small, even for very big networks. Empirical evaluations show that this new measure is correlated with the complexity of the mapping function and thus the generalisation capabilities of the corresponding network. It captures the effective, as oppose to the theoretical, complexity of the network function. We also showcase some uses of the proposed measure for analysis and comparison of trained neural network models.

Juncheng Liu, Steven Mills, Brendan McCane

We introduce a deep recursive octree network for the compression of 3D voxel data. Our network compresses a voxel grid of any size down to a very small latent space in an autoencoder-like network. We show results for compressing 32, 64 and 128 grids down to just 80 floats in the latent space. We demonstrate the effectiveness and efficiency of our proposed method on several publicly available datasets with three experiments: 3D shape classification, 3D shape reconstruction, and shape generation. Experimental results show that our algorithm maintains accuracy while consuming less memory with shorter training times compared to existing methods, especially in 3D reconstruction tasks.

Haitao Xu, Brendan McCane, Lech Szymanski et al.

We show that reinforcement learning agents that learn by surprise (surprisal) get stuck at abrupt environmental transition boundaries because these transitions are difficult to learn. We propose a counter-intuitive solution that we call Mutual Information Minimising Exploration (MIME) where an agent learns a latent representation of the environment without trying to predict the future states. We show that our agent performs significantly better over sharp transition boundaries while matching the performance of surprisal driven agents elsewhere. In particular, we show state-of-the-art performance on difficult learning games such as Gravitar, Montezuma's Revenge and Doom.

Craig Atkinson, Brendan McCane, Lech Szymanski et al.

Pseudo-rehearsal allows neural networks to learn a sequence of tasks without forgetting how to perform in earlier tasks. Preventing forgetting is achieved by introducing a generative network which can produce data from previously seen tasks so that it can be rehearsed along side learning the new task. This has been found to be effective in both supervised and reinforcement learning. Our current work aims to further prevent forgetting by encouraging the generator to accurately generate features important for task retention. More specifically, the generator is improved by introducing a second discriminator into the Generative Adversarial Network which learns to classify between real and fake items from the intermediate activation patterns that they produce when fed through a continual learning agent. Using Atari 2600 games, we experimentally find that improving the generator can considerably reduce catastrophic forgetting compared to the standard pseudo-rehearsal methods used in deep reinforcement learning. Furthermore, we propose normalising the Q-values taught to the long-term system as we observe this substantially reduces catastrophic forgetting by minimising the interference between tasks' reward functions.

Haitao Xu, Brendan McCane, Lech Szymanski

Exploration in environments with continuous control and sparse rewards remains a key challenge in reinforcement learning (RL). Recently, surprise has been used as an intrinsic reward that encourages systematic and efficient exploration. We introduce a new definition of surprise and its RL implementation named Variational Assorted Surprise Exploration (VASE). VASE uses a Bayesian neural network as a model of the environment dynamics and is trained using variational inference, alternately updating the accuracy of the agent's model and policy. Our experiments show that in continuous control sparse reward environments VASE outperforms other surprise-based exploration techniques.

Lech Szymanski, Brendan McCane, Craig Atkinson

We introduce switched linear projections for expressing the activity of a neuron in a deep neural network in terms of a single linear projection in the input space. The method works by isolating the active subnetwork, a series of linear transformations, that determine the entire computation of the network for a given input instance. With these projections we can decompose activity in any hidden layer into patterns detected in a given input instance. We also propose that in ReLU networks it is instructive and meaningful to examine patterns that deactivate the neurons in a hidden layer, something that is implicitly ignored by the existing interpretability methods tracking solely the active aspect of the network's computation.

Tapabrata Chakraborti, Brendan McCane, Steven Mills et al.

Any generic deep machine learning algorithm is essentially a function fitting exercise, where the network tunes its weights and parameters to learn discriminatory features by minimizing some cost function. Though the network tries to learn the optimal feature space, it seldom tries to learn an optimal distance metric in the cost function, and hence misses out on an additional layer of abstraction. We present a simple effective way of achieving this by learning a generic Mahalanabis distance in a collaborative loss function in an end-to-end fashion with any standard convolutional network as the feature learner. The proposed method DML-CRC gives state-of-the-art performance on benchmark fine-grained classification datasets CUB Birds, Oxford Flowers and Oxford-IIIT Pets using the VGG-19 deep network. The method is network agnostic and can be used for any similar classification tasks.

Tapabrata Chakraborti, Brendan McCane, Steven Mills et al.

We present a conditional probabilistic framework for collaborative representation of image patches. It incorporates background compensation and outlier patch suppression into the main formulation itself, thus doing away with the need for pre-processing steps to handle the same. A closed form non-iterative solution of the cost function is derived. The proposed method (PProCRC) outperforms earlier CRC formulations: patch based (PCRC, GP-CRC) as well as the state-of-the-art probabilistic (ProCRC and EProCRC) on three fine-grained species recognition datasets (Oxford Flowers, Oxford-IIIT Pets and CUB Birds) using two CNN backbones (Vgg-19 and ResNet-50).

Tapabrata Chakraborti, Brendan McCane, Steven Mills et al.

We present an end-to-end deep network for fine-grained visual categorization called Collaborative Convolutional Network (CoCoNet). The network uses a collaborative layer after the convolutional layers to represent an image as an optimal weighted collaboration of features learned from training samples as a whole rather than one at a time. This gives CoCoNet more power to encode the fine-grained nature of the data with limited samples. We perform a detailed study of the performance with 1-stage and 2-stage transfer learning. The ablation study shows that the proposed method outperforms its constituent parts consistently. CoCoNet also outperforms few state-of-the-art competing methods. Experiments have been performed on the fine-grained bird species classification problem as a representative example, but the method may be applied to other similar tasks. We also introduce a new public dataset for fine-grained species recognition, that of Indian endemic birds and have reported initial results on it.

Craig Atkinson, Brendan McCane, Lech Szymanski et al.

Neural networks can achieve excellent results in a wide variety of applications. However, when they attempt to sequentially learn, they tend to learn the new task while catastrophically forgetting previous ones. We propose a model that overcomes catastrophic forgetting in sequential reinforcement learning by combining ideas from continual learning in both the image classification domain and the reinforcement learning domain. This model features a dual memory system which separates continual learning from reinforcement learning and a pseudo-rehearsal system that "recalls" items representative of previous tasks via a deep generative network. Our model sequentially learns Atari 2600 games without demonstrating catastrophic forgetting and continues to perform above human level on all three games. This result is achieved without: demanding additional storage requirements as the number of tasks increases, storing raw data or revisiting past tasks. In comparison, previous state-of-the-art solutions are substantially more vulnerable to forgetting on these complex deep reinforcement learning tasks.

Lech Szymanski, Brendan McCane, Michael Albert

We show that the number of unique function mappings in a neural network hypothesis space is inversely proportional to $\prod_lU_l!$, where $U_{l}$ is the number of neurons in the hidden layer $l$.

Brendan McCane, Lech Szymanski

In this paper we introduce new bounds on the approximation of functions in deep networks and in doing so introduce some new deep network architectures for function approximation. These results give some theoretical insight into the success of autoencoders and ResNets.

Craig Atkinson, Brendan McCane, Lech Szymanski et al.

In general, neural networks are not currently capable of learning tasks in a sequential fashion. When a novel, unrelated task is learnt by a neural network, it substantially forgets how to solve previously learnt tasks. One of the original solutions to this problem is pseudo-rehearsal, which involves learning the new task while rehearsing generated items representative of the previous task/s. This is very effective for simple tasks. However, pseudo-rehearsal has not yet been successfully applied to very complex tasks because in these tasks it is difficult to generate representative items. We accomplish pseudo-rehearsal by using a Generative Adversarial Network to generate items so that our deep network can learn to sequentially classify the CIFAR-10, SVHN and MNIST datasets. After training on all tasks, our network loses only 1.67% absolute accuracy on CIFAR-10 and gains 0.24% absolute accuracy on SVHN. Our model's performance is a substantial improvement compared to the current state of the art solution.

Tapabrata Chakraborti, Brendan McCane, Steven Mills et al.

This letter introduces the LOOP binary descriptor (local optimal oriented pattern) that encodes rotation invariance into the main formulation itself. This makes any post processing stage for rotation invariance redundant and improves on both accuracy and time complexity. We consider fine-grained lepidoptera (moth/butterfly) species recognition as the representative problem since it involves repetition of localized patterns and textures that may be exploited for discrimination. We evaluate the performance of LOOP against its predecessors as well as few other popular descriptors. Besides experiments on standard benchmarks, we also introduce a new small image dataset on NZ Lepidoptera. Loop performs as well or better on all datasets evaluated compared to previous binary descriptors. The new dataset and demo code of the proposed method are to be made available through the lead author's academic webpage and GitHub.

Lech Szymanski, Brendan McCane, Wei Gao et al.

Despite being so vital to success of Support Vector Machines, the principle of separating margin maximisation is not used in deep learning. We show that minimisation of margin variance and not maximisation of the margin is more suitable for improving generalisation in deep architectures. We propose the Halfway loss function that minimises the Normalised Margin Variance (NMV) at the output of a deep learning models and evaluate its performance against the Softmax Cross-Entropy loss on the MNIST, smallNORB and CIFAR-10 datasets.

Brendan McCane, Lech Szymanski

We prove that a particular deep network architecture is more efficient at approximating radially symmetric functions than the best known 2 or 3 layer networks. We use this architecture to approximate Gaussian kernel SVMs, and subsequently improve upon them with further training. The architecture and initial weights of the Deep Radial Kernel Network are completely specified by the SVM and therefore sidesteps the problem of empirically choosing an appropriate deep network architecture.

Xiping Fu, Brendan McCane, Steven Mills et al.

Binary codes can be used to speed up nearest neighbor search tasks in large scale data sets as they are efficient for both storage and retrieval. In this paper, we propose a robust auto-encoder model that preserves the geometric relationships of high-dimensional data sets in Hamming space. This is done by considering a noise-removing function in a region surrounding the manifold where the training data points lie. This function is defined with the property that it projects the data points near the manifold into the manifold wisely, and we approximate this function by its first order approximation. Experimental results show that the proposed method achieves better than state-of-the-art results on three large scale high dimensional data sets.