Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userWhy bigger is not always better: on finite and infinite neural networks
202
0
0.0
(
0
)
Authors
Laurence Aitchison
Ask ChatGPT about the research
No Arabic abstract
Recent work has argued that neural networks can be understood theoretically by taking the number of channels to infinity, at which point the outputs become Gaussian process (GP) distributed. However, we note that infinite Bayesian neural networks lack a key facet of the behaviour of real neural networks: the fixed kernel, determined only by network hyperparameters, implies that they cannot do any form of representation learning. The lack of representation or equivalently kernel learning leads to less flexibility and hence worse performance, giving a potential explanation for the inferior performance of infinite networks observed in the literature (e.g. Novak et al. 2019). We give analytic results characterising the prior over representations and representation learning in finite deep linear networks. We show empirically that the representations in SOTA architectures such as ResNets trained with SGD are much closer to those suggested by our deep linear results than by the corresponding infinite network. This motivates the introduction of a new class of network: infinite networks with bottlenecks, which inherit the theoretical tractability of infinite networks while at the same time allowing representation learning.
rate research
Read More
Neural Tangents is a library designed to enable research into infinite-width neural networks. It provides a high-level API for specifying complex and hierarchical neural network architectures. These networks can then be trained and evaluated either at finite-width as usual or in their infinite-width limit. Infinite-width networks can be trained analytically using exact Bayesian inference or using gradient descent via the Neural Tangent Kernel. Additionally, Neural Tangents provides tools to study gradient descent training dynamics of wide but finite networks in either function space or weight space. The entire library runs out-of-the-box on CPU, GPU, or TPU. All computations can be automatically distributed over multiple accelerators with near-linear scaling in the number of devices. Neural Tangents is available at www.github.com/google/neural-tangents. We also provide an accompanying interactive Colab notebook.
Dark matter interacting with the Standard Model fermions through new scalars or pseudoscalars with flavour-diagonal couplings proportional to fermion mass are well motivated theoretically, and provide a useful phenomenological model with which to interpret experimental results. Two modes of dark matter production from these models have been considered in the existing literature: pairs of dark matter produced through top quark loops with an associated monojet in the event, and pair production of dark matter with pairs of heavy flavoured quarks (tops or bottoms). In this paper, we demonstrate that a third, previously overlooked channel yields a non-negligible contribution to LHC dark matter searches in these models. In spite of a generally lower production cross section at LHC when compared to the associated top-pair channel, non-flavour violating single top quark processes are kinematically favored and can significantly increase the sensitivity to these models. Including dark matter production in association with a single top quark through scalar or pseudoscalar mediators, the exclusion limit set by the LHC searches for dark matter can be improved by $30$--$90%$, depending on the mass assumed for the mediator particle.
We provide theoretical and empirical evidence that using tighter evidence lower bounds (ELBOs) can be detrimental to the process of learning an inference network by reducing the signal-to-noise ratio of the gradient estimator. Our results call into question common implicit assumptions that tighter ELBOs are better variational objectives for simultaneous model learning and inference amortization schemes. Based on our insights, we introduce three new algorithms: the partially importance weighted auto-encoder (PIWAE), the multiply importance weighted auto-encoder (MIWAE), and the combination importance weighted auto-encoder (CIWAE), each of which includes the standard importance weighted auto-encoder (IWAE) as a special case. We show that each can deliver improvements over IWAE, even when performance is measured by the IWAE target itself. Furthermore, our results suggest that PIWAE may be able to deliver simultaneous improvements in the training of both the inference and generative networks.
The vast majority of the neural network literature focuses on predicting point values for a given set of response variables, conditioned on a feature vector. In many cases we need to model the full joint conditional distribution over the response variables rather than simply making point predictions. In this paper, we present two novel approaches to such conditional density estimation (CDE): Multiscale Nets (MSNs) and CDE Trend Filtering. Multiscale nets transform the CDE regression task into a hierarchical classification task by decomposing the density into a series of half-spaces and learning boolean probabilities of each split. CDE Trend Filtering applies a k-th order graph trend filtering penalty to the unnormalized logits of a multinomial classifier network, with each edge in the graph corresponding to a neighboring point on a discretized version of the density. We compare both methods against plain multinomial classifier networks and mixture density networks (MDNs) on a simulated dataset and three real-world datasets. The results suggest the two methods are complementary: MSNs work well in a high-data-per-feature regime and CDE-TF is well suited for few-samples-per-feature scenarios where overfitting is a primary concern.
This work is substituted by the paper in arXiv:2011.14066. Stochastic gradient descent is the de facto algorithm for training deep neural networks (DNNs). Despite its popularity, it still requires fine tuning in order to achieve its best performance. This has led to the development of adaptive methods, that claim automatic hyper-parameter optimization. Recently, researchers have studied both algorithmic classes via toy examples: e.g., for over-parameterized linear regression, Wilson et. al. (2017) shows that, while SGD always converges to the minimum-norm solution, adaptive methods show no such inclination, leading to worse generalization capabilities. Our aim is to study this conjecture further. We empirically show that the minimum weight norm is not necessarily the proper gauge of good generalization in simplified scenaria, and different models found by adaptive methods could outperform plain gradient methods. In practical DNN settings, we observe that adaptive methods can outperform SGD, with larger weight norm output models, but without necessarily reducing the amount of tuning required.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
