Yiwei Lyu, Paul Pu Liang, Hai Pham, Eduard Hovy, Barnabás Póczos, Ruslan Salakhutdinov, Louis-Philippe Morency. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Matrix trace estimation is ubiquitous in machine learning applications and has traditionally relied on Hutchinson's method, which requires $O(\log(1/\delta)/\epsilon^2)$ matrix-vector product queries to achieve a $(1 \pm \epsilon)$-multiplicative approximation to $\text{trace}(A)$ with failure probability $\delta$ on positive-semidefinite input matrices $A$. Recently, the Hutch++ algorithm was proposed, which reduces the number of matrix-vector queries from $O(1/\epsilon^2)$ to the optimal $O(1/\epsilon)$, and the algorithm succeeds with constant probability. However, in the high probability setting, the non-adaptive Hutch++ algorithm suffers an extra $O(\sqrt{\log(1/\delta)})$ multiplicative factor in its query complexity. Non-adaptive methods are important, as they correspond to sketching algorithms, which are mergeable, highly parallelizable, and provide low-memory streaming algorithms as well as low-communication distributed protocols. In this work, we close the gap between non-adaptive and adaptive algorithms, showing that even non-adaptive algorithms can achieve $O(\sqrt{\log(1/\delta)}/\epsilon + \log(1/\delta))$ matrix-vector products. In addition, we prove matching lower bounds demonstrating that, up to a $\log \log(1/\delta)$ factor, no further improvement in the dependence on $\delta$ or $\epsilon$ is possible by any non-adaptive algorithm. Finally, our experiments demonstrate the superior performance of our sketch over the adaptive Hutch++ algorithm, which is less parallelizable, as well as over the non-adaptive Hutchinson's method.

Despite the advent of deep learning in computer vision, the general handwriting recognition problem is far from solved. Most existing approaches focus on handwriting datasets that have clearly written text and carefully segmented labels. In this paper, we instead focus on learning handwritten characters from maintenance logs, a constrained setting where data is very limited and noisy. We break the problem into two consecutive stages of word segmentation and word recognition respectively, and utilize data augmentation techniques to train both stages. Extensive comparisons with popular baselines for scene-text detection and word recognition show that our system achieves a lower error rate and is more suited to handle noisy and difficult documents.

We introduce a new scalable approximation for Gaussian processes with provable guarantees which holds simultaneously over its entire parameter space. Our approximation is obtained from an improved sample complexity analysis for sparse spectrum Gaussian processes (SSGPs). In particular, our analysis shows that under a certain data disentangling condition, an SSGP's prediction and model evidence (for training) can well-approximate those of a full GP with low sample complexity. We also develop a new auto-encoding algorithm that finds a latent space to disentangle latent input coordinates into well-separated clusters, which is amenable to our sample complexity analysis. We validate our proposed method on several benchmarks with promising results supporting our theoretical analysis.

Multimodal sentiment analysis is a core research area that studies speaker sentiment expressed from the language, visual, and acoustic modalities. The central challenge in multimodal learning involves inferring joint representations that can process and relate information from these modalities. However, existing work learns joint representations by requiring all modalities as input and as a result, the learned representations may be sensitive to noisy or missing modalities at test time. With the recent success of sequence to sequence (Seq2Seq) models in machine translation, there is an opportunity to explore new ways of learning joint representations that may not require all input modalities at test time. In this paper, we propose a method to learn robust joint representations by translating between modalities. Our method is based on the key insight that translation from a source to a target modality provides a method of learning joint representations using only the source modality as input. We augment modality translations with a cycle consistency loss to ensure that our joint representations retain maximal information from all modalities. Once our translation model is trained with paired multimodal data, we only need data from the source modality at test time for final sentiment prediction. This ensures that our model remains robust from perturbations or missing information in the other modalities. We train our model with a coupled translationprediction objective and it achieves new state-of-the-art results on multimodal sentiment analysis datasets: CMU-MOSI, ICTMMMO, and YouTube. Additional experiments show that our model learns increasingly discriminative joint representations with more input modalities while maintaining robustness to missing or perturbed modalities.

Hadoop, an open-source implementation of MapReduce, is widely used because of its ease of programming, scalability, and availability. With the explosive development of cloud computing, business and scientific applications increasingly take advantage of Hadoop. The sizes of files stored and processed in Hadoop are not bound to very large files anymore. However, Hadoop cannot provide stable and efficient services for small files at both storage and processing levels. To solve these problems, we propose an optimized MapReduce framework for small files, SFMapReduce. In SFMapReduce, we present two techniques, Small File Layout (SFLayout) and customized MapReduce (CMR). SFLayout is used to solve the memory problem and improve I/O performance in HDFS. CMR provides an interface for MapReduce so that SFMapReduce can process MapReduce with SFLayout efficiently. Our experimental results show that SFMapReduce decreases the memory pressure on the Hadoop NameNode, and provides better loading and retrieving throughput. On average, SFMapReduce achieves an improvement on MapReduce processing by 14.5 times and 20.8 times, compared with the original Hadoop and HAR layout.

Integrating energy-harvesting photodiodes with logic and exploiting on-die interconnect capacitance for energy storage can enable new, ultraminiaturized wireless systems. Unlike CMOS imager pixels, the proposed photodiode designs utilize p-diffusion fingers and are implemented in a conventional logic process. Also unlike specialized solar cell processes, the designs utilize the on-chip metal interconnect to form a diffraction grating above the p-diffusion fingers which also provides capacitive energy storage. To explore the tradeoffs between optical efficiency and energy storage for integrated photodiodes, an array of photovoltaics with various diffractive storage capacitors was designed in a 90-nm CMOS logic process. The diffractive effects can be exploited to increase the photodiodes' response to off-axis illumination. Transient effects from interfacing the photodiodes with switched-capacitor DC-DC converters were examined, with measurements indicating a 50% reduction in the output voltage ripple due to the diffractive storage capacitance. A quantitative comparison between 90-nm and 0.35-μm CMOS logic processes for energy-harvesting capabilities was carried out. Measurements show an increase in power generation for the newer CMOS technology, however at the cost of reduced output voltage. One potential application for the integrated photodiodes is harvesting energy for a subdermal biomedical device.

As the density of embedded memory increases, manufacturing yields of integrated circuits can reach unacceptable limits. Normal memory testing operations require BIST to effectively deal with problems such as limited access and "at speed" testing. In this paper we describe a novel methodology that extends the BIST concept to diagnosis and repair utilizing redundant components. We describe an application using redundant columns and accompanying algorithms. It allows for the autonomous repair of defective circuitry without external stimulus (e.g. laser repair). The method has been implemented with negligible timing penalties and reasonable area overhead.