Mehul Motani

John Chong Min Tan, Prince Saroj, Bharat Runwal et al.

TaskGen is an open-sourced agentic framework which uses an Agent to solve an arbitrary task by breaking them down into subtasks. Each subtask is mapped to an Equipped Function or another Agent to execute. In order to reduce verbosity (and hence token usage), TaskGen uses StrictJSON that ensures JSON output from the Large Language Model (LLM), along with additional features such as type checking and iterative error correction. Key to the philosophy of TaskGen is the management of information/memory on a need-to-know basis. We empirically evaluate TaskGen on various environments such as 40x40 dynamic maze navigation with changing obstacle locations (100% solve rate), TextWorld escape room solving with dense rewards and detailed goals (96% solve rate), web browsing (69% of actions successful), solving the MATH dataset (71% solve rate over 100 Level-5 problems), Retrieval Augmented Generation on NaturalQuestions dataset (F1 score of 47.03%)

John Tan Chong Min, Mehul Motani

Modern deep neural networks require a significant amount of computing time and power to train and deploy, which limits their usage on edge devices. Inspired by the iterative weight pruning in the Lottery Ticket Hypothesis, we propose DropNet, an iterative pruning method which prunes nodes/filters to reduce network complexity. DropNet iteratively removes nodes/filters with the lowest average post-activation value across all training samples. Empirically, we show that DropNet is robust across diverse scenarios, including MLPs and CNNs using the MNIST, CIFAR-10 and Tiny ImageNet datasets. We show that up to 90% of the nodes/filters can be removed without any significant loss of accuracy. The final pruned network performs well even with reinitialization of the weights and biases. DropNet also has similar accuracy to an oracle which greedily removes nodes/filters one at a time to minimise training loss, highlighting its effectiveness.

Shiyu Liu, Mehul Motani

Mutual Information (MI) based feature selection makes use of MI to evaluate each feature and eventually shortlists a relevant feature subset, in order to address issues associated with high-dimensional datasets. Despite the effectiveness of MI in feature selection, we notice that many state-of-the-art algorithms disregard the so-called unique relevance (UR) of features, and arrive at a suboptimal selected feature subset which contains a non-negligible number of redundant features. We point out that the heart of the problem is that all these MIBFS algorithms follow the criterion of Maximize Relevance with Minimum Redundancy (MRwMR), which does not explicitly target UR. This motivates us to augment the existing criterion with the objective of boosting unique relevance (BUR), leading to a new criterion called MRwMR-BUR. Depending on the task being addressed, MRwMR-BUR has two variants, termed MRwMR-BUR-KSG and MRwMR-BUR-CLF, which estimate UR differently. MRwMR-BUR-KSG estimates UR via a nearest-neighbor based approach called the KSG estimator and is designed for three major tasks: (i) Classification Performance. (ii) Feature Interpretability. (iii) Classifier Generalization. MRwMR-BUR-CLF estimates UR via a classifier based approach. It adapts UR to different classifiers, further improving the competitiveness of MRwMR-BUR for classification performance oriented tasks. The performance of both MRwMR-BUR-KSG and MRwMR-BUR-CLF is validated via experiments using six public datasets and three popular classifiers. Specifically, as compared to MRwMR, the proposed MRwMR-BUR-KSG improves the test accuracy by 2% - 3% with 25% - 30% fewer features being selected, without increasing the algorithm complexity. MRwMR-BUR-CLF further improves the classification performance by 3.8%- 5.5% (relative to MRwMR), and it also outperforms three popular classifier dependent feature selection methods.

John Chong Min Tan, Mehul Motani

We attempt to solve the Abstraction and Reasoning Corpus (ARC) Challenge using Large Language Models (LLMs) as a system of multiple expert agents. Using the flexibility of LLMs to be prompted to do various novel tasks using zero-shot, few-shot, context-grounded prompting, we explore the feasibility of using LLMs to solve the ARC Challenge. We firstly convert the input image into multiple suitable text-based abstraction spaces. We then utilise the associative power of LLMs to derive the input-output relationship and map this to actions in the form of a working program, similar to Voyager / Ghost in the MineCraft. In addition, we use iterative environmental feedback in order to guide LLMs to solve the task. Our proposed approach achieves 50 solves out of 111 training set problems (45%) with just three abstraction spaces - grid, object and pixel - and we believe that with more abstraction spaces and learnable actions, we will be able to solve more.

Shiyu Liu, Rohan Ghosh, Mehul Motani

Long-range time series forecasting is usually based on one of two existing forecasting strategies: Direct Forecasting and Iterative Forecasting, where the former provides low bias, high variance forecasts and the latter leads to low variance, high bias forecasts. In this paper, we propose a new forecasting strategy called Generative Forecasting (GenF), which generates synthetic data for the next few time steps and then makes long-range forecasts based on generated and observed data. We theoretically prove that GenF is able to better balance the forecasting variance and bias, leading to a much smaller forecasting error. We implement GenF via three components: (i) a novel conditional Wasserstein Generative Adversarial Network (GAN) based generator for synthetic time series data generation, called CWGAN-TS. (ii) a transformer based predictor, which makes long-range predictions using both generated and observed data. (iii) an information theoretic clustering algorithm to improve the training of both the CWGAN-TS and the transformer based predictor. The experimental results on five public datasets demonstrate that GenF significantly outperforms a diverse range of state-of-the-art benchmarks and classical approaches. Specifically, we find a 5% - 11% improvement in predictive performance (mean absolute error) while having a 15% - 50% reduction in parameters compared to the benchmarks. Lastly, we conduct an ablation study to further explore and demonstrate the effectiveness of the components comprising GenF.

Rohan Ghosh, Mehul Motani

Most entropy measures depend on the spread of the probability distribution over the sample space $\mathcal{X}$, and the maximum entropy achievable scales proportionately with the sample space cardinality $|\mathcal{X}|$. For a finite $|\mathcal{X}|$, this yields robust entropy measures which satisfy many important properties, such as invariance to bijections, while the same is not true for continuous spaces (where $|\mathcal{X}|=\infty$). Furthermore, since $\mathbb{R}$ and $\mathbb{R}^d$ ($d\in \mathbb{Z}^+$) have the same cardinality (from Cantor's correspondence argument), cardinality-dependent entropy measures cannot encode the data dimensionality. In this work, we question the role of cardinality and distribution spread in defining entropy measures for continuous spaces, which can undergo multiple rounds of transformations and distortions, e.g., in neural networks. We find that the average value of the local intrinsic dimension of a distribution, denoted as ID-Entropy, can serve as a robust entropy measure for continuous spaces, while capturing the data dimensionality. We find that ID-Entropy satisfies many desirable properties and can be extended to conditional entropy, joint entropy and mutual-information variants. ID-Entropy also yields new information bottleneck principles and also links to causality. In the context of deep learning, for feedforward architectures, we show, theoretically and empirically, that the ID-Entropy of a hidden layer directly controls the generalization gap for both classifiers and auto-encoders, when the target function is Lipschitz continuous. Our work primarily shows that, for continuous spaces, taking a structural rather than a statistical approach yields entropy measures which preserve intrinsic data dimensionality, while being relevant for studying various architectures.

John Chong Min Tan, Mehul Motani

Model-based next state prediction and state value prediction are slow to converge. To address these challenges, we do the following: i) Instead of a neural network, we do model-based planning using a parallel memory retrieval system (which we term the slow mechanism); ii) Instead of learning state values, we guide the agent's actions using goal-directed exploration, by using a neural network to choose the next action given the current state and the goal state (which we term the fast mechanism). The goal-directed exploration is trained online using hippocampal replay of visited states and future imagined states every single time step, leading to fast and efficient training. Empirical studies show that our proposed method has a 92% solve rate across 100 episodes in a dynamically changing grid world, significantly outperforming state-of-the-art actor critic mechanisms such as PPO (54%), TRPO (50%) and A2C (24%). Ablation studies demonstrate that both mechanisms are crucial. We posit that the future of Reinforcement Learning (RL) will be to model goals and sub-goals for various tasks, and plan it out in a goal-directed memory-based approach.

John Tan Chong Min, Mehul Motani

Traditional reinforcement learning (RL) environments typically are the same for both the training and testing phases. Hence, current RL methods are largely not generalizable to a test environment which is conceptually similar but different from what the method has been trained on, which we term the novel test environment. As an effort to push RL research towards algorithms which can generalize to novel test environments, we introduce the Brick Tic-Tac-Toe (BTTT) test bed, where the brick position in the test environment is different from that in the training environment. Using a round-robin tournament on the BTTT environment, we show that traditional RL state-search approaches such as Monte Carlo Tree Search (MCTS) and Minimax are more generalizable to novel test environments than AlphaZero is. This is surprising because AlphaZero has been shown to achieve superhuman performance in environments such as Go, Chess and Shogi, which may lead one to think that it performs well in novel test environments. Our results show that BTTT, though simple, is rich enough to explore the generalizability of AlphaZero. We find that merely increasing MCTS lookahead iterations was insufficient for AlphaZero to generalize to some novel test environments. Rather, increasing the variety of training environments helps to progressively improve generalizability across all possible starting brick configurations.

Shiyu Liu, Rohan Ghosh, Dylan Tan et al.

The rectified linear unit (ReLU) is a highly successful activation function in neural networks as it allows networks to easily obtain sparse representations, which reduces overfitting in overparameterized networks. However, in network pruning, we find that the sparsity introduced by ReLU, which we quantify by a term called dynamic dead neuron rate (DNR), is not beneficial for the pruned network. Interestingly, the more the network is pruned, the smaller the dynamic DNR becomes during optimization. This motivates us to propose a method to explicitly reduce the dynamic DNR for the pruned network, i.e., de-sparsify the network. We refer to our method as Activating-while-Pruning (AP). We note that AP does not function as a stand-alone method, as it does not evaluate the importance of weights. Instead, it works in tandem with existing pruning methods and aims to improve their performance by selective activation of nodes to reduce the dynamic DNR. We conduct extensive experiments using popular networks (e.g., ResNet, VGG) via two classical and three state-of-the-art pruning methods. The experimental results on public datasets (e.g., CIFAR-10/100) suggest that AP works well with existing pruning methods and improves the performance by 3% - 4%. For larger scale datasets (e.g., ImageNet) and state-of-the-art networks (e.g., vision transformer), we observe an improvement of 2% - 3% with AP as opposed to without. Lastly, we conduct an ablation study to examine the effectiveness of the components comprising AP.

Shiyu Liu, Rohan Ghosh, John Tan Chong Min et al.

The importance of learning rate (LR) schedules on network pruning has been observed in a few recent works. As an example, Frankle and Carbin (2019) highlighted that winning tickets (i.e., accuracy preserving subnetworks) can not be found without applying a LR warmup schedule and Renda, Frankle and Carbin (2020) demonstrated that rewinding the LR to its initial state at the end of each pruning cycle improves performance. In this paper, we go one step further by first providing a theoretical justification for the surprising effect of LR schedules. Next, we propose a LR schedule for network pruning called SILO, which stands for S-shaped Improved Learning rate Optimization. The advantages of SILO over existing state-of-the-art (SOTA) LR schedules are two-fold: (i) SILO has a strong theoretical motivation and dynamically adjusts the LR during pruning to improve generalization. Specifically, SILO increases the LR upper bound (max_lr) in an S-shape. This leads to an improvement of 2% - 4% in extensive experiments with various types of networks (e.g., Vision Transformers, ResNet) on popular datasets such as ImageNet, CIFAR-10/100. (ii) In addition to the strong theoretical motivation, SILO is empirically optimal in the sense of matching an Oracle, which exhaustively searches for the optimal value of max_lr via grid search. We find that SILO is able to precisely adjust the value of max_lr to be within the Oracle optimized interval, resulting in performance competitive with the Oracle with significantly lower complexity.

Tianci Zhang, Aobo Liu, Zhengchuan Chen et al.

This paper studies information timeliness in multi-source wireless Internet of Things (IoT) status updating systems under a truncated Automatic Repeat reQuest (ARQ) protocol. We propose a source-aware truncated ARQ (SATARQ) scheme that allows differentiated maximum transmission times (MTTs) tailored to different sources. This work focuses on a wireless system with preemptive update management. To study the statistical characteristics of the age of information (AoI) process for each source, a multi-dimensional age process (MDAP) is developed and modeled as a Markov chain, tracking both the AoI and the age of the concerned source's update currently in transmission. Via Markov analysis of the MDAP, we obtain analytical expressions for the distributions and averages of the AoI and peak AoI, as well as the average power consumption of IoT device. The timeliness-energy tradeoff is analyzed by examining the impact of the MTT, update generation probability (UGP), and wireless transmission power (TP). Moreover, this work explores the energy efficiency of the wireless status updating process and its relationship with the information timeliness and energy cost. Numerical results validate the theoretical analysis. Finally, it is demonstrated that the proposed SATARQ, combined with the optimization of MTTs, UGPs, and TPs, significantly improves the overall timeliness-energy tradeoff and energy efficiency across all sources.

Yunze Leng, Rohan Ghosh, Mehul Motani

Supervised learning with tabular data presents unique challenges, including low data sizes, the absence of structural cues, and heterogeneous features spanning both categorical and continuous domains. Unlike vision and language tasks, where models can exploit inductive biases in the data, tabular data lacks inherent positional structure, hindering the effectiveness of self-attention mechanisms. While recent transformer-based models like TabTransformer, SAINT, and FT-Transformer (which we refer to as 3T) have shown promise on tabular data, they typically operate without leveraging structural cues such as positional encodings (PEs), as no prior structural information is usually available. In this work, we find both theoretically and empirically that structural cues, specifically PEs can be a useful tool to improve generalization performance for tabular transformers. We find that PEs impart the ability to reduce the effective rank (a form of intrinsic dimensionality) of the features, effectively simplifying the task by reducing the dimensionality of the problem, yielding improved generalization. To that end, we propose Tab-PET (PEs for Tabular Transformers), a graph-based framework for estimating and inculcating PEs into embeddings. Inspired by approaches that derive PEs from graph topology, we explore two paradigms for graph estimation: association-based and causality-based. We empirically demonstrate that graph-derived PEs significantly improve performance across 50 classification and regression datasets for 3T. Notably, association-based graphs consistently yield more stable and pronounced gains compared to causality-driven ones. Our work highlights an unexpected role of PEs in tabular transformers, revealing how they can be harnessed to improve generalization.

Soumyadeep Dhar, Kei Sen Fong, Mehul Motani

Distilling large neural networks into simple, human-readable symbolic formulas is a promising path toward trustworthy and interpretable AI. However, this process is often brittle, as the complex functions learned by standard networks are poor targets for symbolic discovery, resulting in low-fidelity student models. In this work, we propose a novel training paradigm to address this challenge. Instead of passively distilling a pre-trained network, we introduce a \textbf{Jacobian-based regularizer} that actively encourages the ``teacher'' network to learn functions that are not only accurate but also inherently smoother and more amenable to distillation. We demonstrate through extensive experiments on a suite of real-world regression benchmarks that our method is highly effective. By optimizing the regularization strength for each problem, we improve the $R^2$ score of the final distilled symbolic model by an average of \textbf{120\% (relative)} compared to the standard distillation pipeline, all while maintaining the teacher's predictive accuracy. Our work presents a practical and principled method for significantly improving the fidelity of interpretable models extracted from complex neural networks.

Vikrant Malik, Rohan Ghosh, Mehul Motani

The advancement of deep learning has led to the development of neural decoders for low latency communications. However, neural decoders can be very complex which can lead to increased computation and latency. We consider iterative pruning approaches (such as the lottery ticket hypothesis algorithm) to prune weights in neural decoders. Decoders with fewer number of weights can have lower latency and lower complexity while retaining the accuracy of the original model. This will make neural decoders more suitable for mobile and other edge devices with limited computational power. We also propose semi-soft decision decoding for neural decoders which can be used to improve the bit error rate performance of the pruned network.

Shiyu Liu, Rohan Ghosh, Mehul Motani

Long-range time series forecasting is usually based on one of two existing forecasting strategies: Direct Forecasting and Iterative Forecasting, where the former provides low bias, high variance forecasts and the later leads to low variance, high bias forecasts. In this paper, we propose a new forecasting strategy called Generative Forecasting (GenF), which generates synthetic data for the next few time steps and then makes long-range forecasts based on generated and observed data. We theoretically prove that GenF is able to better balance the forecasting variance and bias, leading to a much smaller forecasting error. We implement GenF via three components: (i) a novel conditional Wasserstein Generative Adversarial Network (GAN) based generator for synthetic time series data generation, called CWGAN-TS. (ii) a transformer based predictor, which makes long-range predictions using both generated and observed data. (iii) an information theoretic clustering algorithm to improve the training of both the CWGAN-TS and the transformer based predictor. The experimental results on five public datasets demonstrate that GenF significantly outperforms a diverse range of state-of-the-art benchmarks and classical approaches. Specifically, we find a 5% - 11% improvement in predictive performance (mean absolute error) while having a 15% - 50% reduction in parameters compared to the benchmarks. Lastly, we conduct an ablation study to demonstrate the effectiveness of the components comprising GenF.

Shiyu Liu, Chong Min John Tan, Mehul Motani

We explore a new perspective on adapting the learning rate (LR) schedule to improve the performance of the ReLU-based network as it is iteratively pruned. Our work and contribution consist of four parts: (i) We find that, as the ReLU-based network is iteratively pruned, the distribution of weight gradients tends to become narrower. This leads to the finding that as the network becomes more sparse, a larger value of LR should be used to train the pruned network. (ii) Motivated by this finding, we propose a novel LR schedule, called S-Cyclical (S-Cyc) which adapts the conventional cyclical LR schedule by gradually increasing the LR upper bound (max_lr) in an S-shape as the network is iteratively pruned.We highlight that S-Cyc is a method agnostic LR schedule that applies to many iterative pruning methods. (iii) We evaluate the performance of the proposed S-Cyc and compare it to four LR schedule benchmarks. Our experimental results on three state-of-the-art networks (e.g., VGG-19, ResNet-20, ResNet-50) and two popular datasets (e.g., CIFAR-10, ImageNet-200) demonstrate that S-Cyc consistently outperforms the best performing benchmark with an improvement of 2.1% - 3.4%, without substantial increase in complexity. (iv) We evaluate S-Cyc against an oracle and show that S-Cyc achieves comparable performance to the oracle, which carefully tunes max_lr via grid search.

Shiyu Liu, Mehul Motani

Vital signs including heart rate, respiratory rate, body temperature and blood pressure, are critical in the clinical decision making process. Effective early prediction of vital signs help to alert medical practitioner ahead of time and may prevent adverse health outcomes. In this paper, we suggest a new approach called generative boosting, in order to effectively perform early prediction of vital signs. Generative boosting consists of a generative model, to generate synthetic data for next few time steps, and several predictive models, to directly make long-range predictions based on observed and generated data. We explore generative boosting via long short-term memory (LSTM) for both the predictive and generative models, leading to a scheme called generative LSTM (GLSTM). Our experiments indicate that GLSTM outperforms a diverse range of strong benchmark models, with and without generative boosting. Finally, we use a mutual information based clustering algorithm to select a more representative dataset to train the generative model of GLSTM. This significantly improves the long-range predictive performance of high variation vital signs such as heart rate and systolic blood pressure.

Rohan Ghosh, Anupam K. Gupta, Mehul Motani

Convolutional Neural Networks (CNN) have been pivotal to the success of many state-of-the-art classification problems, in a wide variety of domains (for e.g. vision, speech, graphs and medical imaging). A commonality within those domains is the presence of hierarchical, spatially agglomerative local-to-global interactions within the data. For two-dimensional images, such interactions may induce an a priori relationship between the pixel data and the underlying spatial ordering of the pixels. For instance in natural images, neighboring pixels are more likely contain similar values than non-neighboring pixels which are further apart. To that end, we propose a statistical metric called spatial orderness, which quantifies the extent to which the input data (2D) obeys the underlying spatial ordering at various scales. In our experiments, we mainly find that adding convolutional layers to a CNN could be counterproductive for data bereft of spatial order at higher scales. We also observe, quite counter-intuitively, that the spatial orderness of CNN feature maps show a synchronized increase during the intial stages of training, and validation performance only improves after spatial orderness of feature maps start decreasing. Lastly, we present a theoretical analysis (and empirical validation) of the spatial orderness of network weights, where we find that using smaller kernel sizes leads to kernels of greater spatial orderness and vice-versa.

Shiyu Liu, Mehul Motani

Feature selection, which searches for the most representative features in observed data, is critical for health data analysis. Unlike feature extraction, such as PCA and autoencoder based methods, feature selection preserves interpretability, meaning that the selected features provide direct information about certain health conditions (i.e., the label). Thus, feature selection allows domain experts, such as clinicians, to understand the predictions made by machine learning based systems, as well as improve their own diagnostic skills. Mutual information is often used as a basis for feature selection since it measures dependencies between features and labels. In this paper, we introduce a novel mutual information based feature selection (MIBFS) method called SURI, which boosts features with high unique relevant information. We compare SURI to existing MIBFS methods using 3 different classifiers on 6 publicly available healthcare data sets. The results indicate that, in addition to preserving interpretability, SURI selects more relevant feature subsets which lead to higher classification performance. More importantly, we explore the dynamics of mutual information on a public low-dimensional health data set via exhaustive search. The results suggest the important role of unique relevant information in feature selection and verify the principles behind SURI.

Lin Zhou, Vincent Y. F. Tan, Mehul Motani

Motivated by real-world machine learning applications, we analyze approximations to the non-asymptotic fundamental limits of statistical classification. In the binary version of this problem, given two training sequences generated according to two {\em unknown} distributions $P_1$ and $P_2$, one is tasked to classify a test sequence which is known to be generated according to either $P_1$ or $P_2$. This problem can be thought of as an analogue of the binary hypothesis testing problem but in the present setting, the generating distributions are unknown. Due to finite sample considerations, we consider the second-order asymptotics (or dispersion-type) tradeoff between type-I and type-II error probabilities for tests which ensure that (i) the type-I error probability for {\em all} pairs of distributions decays exponentially fast and (ii) the type-II error probability for a {\em particular} pair of distributions is non-vanishing. We generalize our results to classification of multiple hypotheses with the rejection option.