We consider the problem of length generalization in sequence prediction. We define a new metric of performance in this setting – the Asymmetric-Regret– which measures regret against a benchmark predictor with longer context length than available to the learner. We continue by studying this concept through the lens of the spectral filter-ing algorithm. We present a gradient-based learn-ing algorithm that provably achieves length generalization for linear dynamical systems. We conclude with proof-of-concept experiments which are consistent with our theory.

We study the problem of preconditioning in the setting of sequential prediction. From the theoretical lens of linear dynamical systems, we show that applying a convolution to the input sequence translates to applying a polynomial to the unknown transition matrix in the hidden space. With this insight, we develop a novel preconditioning method that convolves the input sequence with the coefficients of the Chebyshev or Legendre polynomials. We formally prove that this improves the regret of a wide family of prediction methods. We proceed to apply this preconditioning technique to the method of spectral filtering. This gives the first sublinear regret bound that is also hidden-dimension free (up to logarithmic factors) even when the hidden transition matrix is asymmetric. From rigorous experiments on synthetic data we show that our simple preconditioning method generalizes to both 1) settings where the data is \emph{not} from a linear dynamical system and 2) a broad range of learning algorithms, including recurrent neural networks.

We consider the problem of minimizing a $d$-dimensional Lipschitz convex function using a stochastic gradient oracle. We introduce and motivate a setting where the noise of the stochastic gradient is isotropic in that it is bounded in every direction with high probability. We then develop an algorithm for this setting which improves upon prior results by a factor of $d$ in certain regimes, and as a corollary, achieves a new state-of-the-art complexity for sub-exponential noise. We give matching lower bounds (up to polylogarithmic factors) for both results. Additionally, we develop an efficient quantum isotropifier, a quantum algorithm which converts a variance-bounded quantum sampling oracle into one that outputs an unbiased estimate with isotropic error. Combining our results, we obtain improved dimension-dependent rates for quantum stochastic convex optimization.

We show that any memory-constrained, first-order algorithm which minimizesd-dimensional, 1-Lipschitz convex functions over the unit ball to 1/poly(d) accuracy using at most d1.25 - δbits of memory must make at least \(\tilde{\Omega }(d^{1 + (4/3)\delta })\) first-order queries (for any constant \(\delta \in [0, 1/4]\) ). Consequently, the performance of such memory-constrained algorithms are at least a polynomial factor worse than the optimal Õ(d) query bound for this problem obtained by cutting plane methods that use Õ(d2) memory. This resolves one of the open problems in the COLT 2019 open problem publication of Woodworth and Srebro.

Minimizing a convex function with access to a first order oracle---that returns the function evaluation and (sub)gradient at a query point---is a canonical optimization problem and a fundamental primitive in machine learning. Gradient-based methods are the most popular approaches used for solving the problem, owing to their simplicity and computational efficiency. These methods, however, do not achieve the information-theoretically optimal query complexity for minimizing the underlying function to small error, which are achieved by more expensive techniques based on cutting-plane methods. Is it possible to achieve the information-theoretically query complexity without using these more complex and computationally expensive methods? In this work, we use memory as a lens to understand this, and show that is is not possible to achieve optimal query complexity without using significantly more memory than that used by gradient descent.

We provide new gradient-based methods for efficiently solving a broad class of ill-conditioned optimization problems. We consider the problem of minimizing a function $f : \mathbb{R}^d \rightarrow \mathbb{R}$ which is implicitly decomposable as the sum of $m$ unknown non-interacting smooth, strongly convex functions and provide a method which solves this problem with a number of gradient evaluations that scales (up to logarithmic factors) as the product of the square-root of the condition numbers of the components. This complexity bound (which we prove is nearly optimal) can improve almost exponentially on that of accelerated gradient methods, which grow as the square root of the condition number of $f$. Additionally, we provide efficient methods for solving stochastic, quadratic variants of this multiscale optimization problem. Rather than learn the decomposition of $f$ (which would be prohibitively expensive), our methods apply a clean recursive “Big-Step-Little-Step” interleaving of standard methods. The resulting algorithms use $\tilde{\mathcal{O}}(d m)$ space, are numerically stable, and open the door to a more fine-grained understanding of the complexity of convex optimization beyond condition number.

We show that any memory-constrained, first-order algorithm which minimizes $d$-dimensional, $1$-Lipschitz convex functions over the unit ball to $1/\mathrm{poly}(d)$ accuracy using at most $d^{1.25 - \delta}$ bits of memory must make at least $\Omega(d^{1 + (4/3)\delta})$ first-order queries (for any constant $\delta \in [0, 1/4]$). Consequently, the performance of such memory-constrained algorithms are a polynomial factor worse than the optimal $\tilde{O}(d)$ query bound for this problem obtained by cutting plane methods that use $\tilde{O}(d^2)$ memory. This resolves one of the open problems in the COLT 2019 open problem publication of Woodworth and Srebro.

We study probabilistic prediction games when the underlying model is misspecified, investigating the consequences of predicting using an incorrect parametric model. We show that for a broad class of loss functions and parametric families of distributions, the regret of playing a “proper” predictor—one from the putative model class—relative to the best predictor in the same model class has lower bound scaling at least as $\sqrt{\gamma n}$, where $\gamma$ is a measure of the model misspecification to the true distribution in terms of total variation distance. In contrast, using an aggregation-based (improper) learner, one can obtain regret $d \log n$ for any underlying generating distribution, where $d$ is the dimension of the parameter; we exhibit instances in which this is unimprovable even over the family of all learners that may play distributions in the convex hull of the parametric family. These results suggest that simple strategies for aggregating multiple learners together should be more robust, and several experiments conform to this hypothesis.