SpectralLeader: Online Spectral Learning for Single Topic Models
This provides a globally convergent alternative to online EM for learning latent variable models in streaming settings, addressing a bottleneck in online learning for practitioners.
The authors tackled the problem of learning latent variable models from streaming data by developing SpectralLeader, an online spectral learning algorithm that guarantees convergence to the global optimum and achieves sublinear regret, performing similarly to or better than online EM in experiments.
We study the problem of learning a latent variable model from a stream of data. Latent variable models are popular in practice because they can explain observed data in terms of unobserved concepts. These models have been traditionally studied in the offline setting. In the online setting, on the other hand, the online EM is arguably the most popular algorithm for learning latent variable models. Although the online EM is computationally efficient, it typically converges to a local optimum. In this work, we develop a new online learning algorithm for latent variable models, which we call SpectralLeader. SpectralLeader always converges to the global optimum, and we derive a sublinear upper bound on its $n$-step regret in the bag-of-words model. In both synthetic and real-world experiments, we show that SpectralLeader performs similarly to or better than the online EM with tuned hyper-parameters.