Kinjal Basu, Ibrahim Abdelaziz, Kiran Kate, Mayank Agarwal, Maxwell Crouse, Yara Rizk, Kelsey Bradford, Asim Munawar, Sadhana Kumaravel, Saurabh Goyal, Xin Wang, Luis A. Lastras, Pavan Kapanipathi. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

CNNs are used extensively for computer vision tasks like activity recognition, image classification, segmentation etc. The large compute memory required in these applications restricts the use of high batch size during inference, thereby increasing the overall prediction time. Prior work addresses this issue through various model compression mechanisms like weight/filter pruning, quantizing the parameters/intermediate outputs, etc. We propose a complementary technique where we improve inference time by using variable batch sizes (VBS) across the layers of a CNN. This optimises the memory-time trade-off for each layer and leads to better network throughput. Our approach does not make any modifications to the existing network (unlike pruning or quantization techniques) and thus there is no impact on the model accuracy. We develop a dynamic program (DP) based algorithm that takes inference time and memory required by different layers of the network as input, and computes the optimal batch sizes for each layer depending on the available resources (RAM, storage space etc.). We demonstrate our findings in two different settings: video inference on K80 GPUs and image inference on Edge devices. On video networks like C3D, our VBS algorithm gives up to 61% higher throughput compared to a fixed batch size baseline. On image networks like GoogleNet, ResNet50 etc., we achieve up to 60% higher throughput compared to a fixed batch size baseline.

We develop a novel method, called PoWER-BERT, for improving the inference time of the popular BERT model, while maintaining the accuracy. It works by: a) exploiting redundancy pertaining to word-vectors (intermediate transformer block outputs) and eliminating the redundant vectors. b) determining which word-vectors to eliminate by developing a strategy for measuring their significance, based on the self-attention mechanism. c) learning how many word-vectors to eliminate by augmenting the BERT model and the loss function. Experiments on the standard GLUE benchmark shows that PoWER-BERT achieves up to 4.5x reduction in inference time over BERT with < 1% loss in accuracy. We show that PoWER-BERT offers significantly better trade-off between accuracy and inference time compared to prior methods. We demonstrate that our method attains up to 6.8x reduction in inference time with < 1% loss in accuracy when applied over ALBERT, a highly compressed version of BERT. The code for PoWER-BERT is publicly available at https://github.com/IBM/PoWER-BERT.

Several real-world applications require real-time prediction on resource-scarce devices such as an Internet of Things (IoT) sensor. Such applications demand prediction models with small storage and computational complexity that do not compromise significantly on accuracy. In this work, we propose ProtoNN, a novel algorithm that addresses the problem of real-time and accurate prediction on resource-scarce devices. ProtoNN is inspired by k-Nearest Neighbor (KNN) but has several orders lower storage and prediction complexity. ProtoNN models can be deployed even on devices with puny storage and computational power (e.g. an Arduino UNO with 2kB RAM) to get excellent prediction accuracy. ProtoNN derives its strength from three key ideas: a) learning a small number of prototypes to represent the entire training set, b) sparse low dimensional projection of data, c) joint discriminative learning of the projection and prototypes with explicit model size constraint. We conduct systematic empirical evaluation of ProtoNN on a variety of supervised learning tasks (binary, multi-class, multi-label classification) and show that it gives nearly state-of-the-art prediction accuracy on resource-scarce devices while consuming several orders lower storage, and using minimal working memory.

This paper develops a novel tree-based algorithm, called Bonsai, for efficient prediction on IoT devices – such as those based on the Arduino Uno board having an 8 bit ATmega328P microcontroller operating at 16 MHz with no native floating point support, 2 KB RAM and 32 KB read-only flash. Bonsai maintains prediction accuracy while minimizing model size and prediction costs by: (a) developing a tree model which learns a single, shallow, sparse tree with powerful nodes; (b) sparsely projecting all data into a low-dimensional space in which the tree is learnt; and (c) jointly learning all tree and projection parameters. Experimental results on multiple benchmark datasets demonstrate that Bonsai can make predictions in milliseconds even on slow microcontrollers, can fit in KB of memory, has lower battery consumption than all other algorithms while achieving prediction accuracies that can be as much as 30\% higher than state-of-the-art methods for resource-efficient machine learning. Bonsai is also shown to generalize to other resource constrained settings beyond IoT by generating significantly better search results as compared to Bing’s L3 ranker when the model size is restricted to 300 bytes. Bonsai’s code can be downloaded from (http://www.manikvarma.org/code/Bonsai/download.html).

The evidence-based analysis of people's navigation and wayfinding behaviour in large-scale built-up environments (e.g., hospitals, airports) encompasses the measurement and qualitative analysis of a range of aspects including people's visual perception in new and familiar surroundings, their decision-making procedures and intentions, the affordances of the environment itself, etc. In our research on large-scale evidence-based qualitative analysis of wayfinding behaviour, we construe visual perception and navigation in built-up environments as a dynamic narrative construction process of movement and exploration driven by situation-dependent goals, guided by visual aids such as signage and landmarks, and influenced by environmental (e.g., presence of other people, time of day, lighting) and personal (e.g., age, physical attributes) factors. We employ a range of sensors for measuring the embodied visuo-locomotive experience of building users: eye-tracking, egocentric gaze analysis, external camera based visual analysis to interpret fine-grained behaviour (e.g., stopping, looking around, interacting with other people), and also manual observations made by human experimenters. Observations are processed, analysed, and integrated in a holistic model of the visuo-locomotive narrative experience at the individual and group level. Our model also combines embodied visual perception analysis with analysis of the structure and layout of the environment (e.g., topology, routes, isovists) computed from available 3D models of the building. In this framework, abstract regions like the visibility space, regions of attention, eye movement clusters, are treated as first class visuo-spatial and iconic objects that can be used for interpreting the visual experience of subjects in a high-level qualitative manner. The final integrated analysis of the wayfinding experience is such that it can even be presented in a virtual reality environment thereby providing an immersive experience (e.g., using tools such as the Oculus Rift) of the qualitative analysis for single participants, as well as for a combined analysis of large group. This capability is especially important for experiments in post-occupancy analysis of building performance. Our construction of indoor wayfinding experience as a form of moving image analysis centralizes the role and influence of perceptual visuo-spatial characteristics and morphological features of the built environment into the discourse on wayfinding research. We will demonstrate the impact of this work with several case-studies, particularly focussing on a large-scale experiment conducted at the New Parkland Hospital in Dallas Texas, USA.