Publications

Also on Google Scholar.

Thesis

Granted US patents

• 2022-07
  Goal Oriented Dialog Generation using Dialog Template, API and Entity data

  Anish Acharya, A Metallinou, T Chung, S Paul, S Chandra, C Lin, D H Tur, A Mandal

  US11393454 · Amazon · USPTO
• 2020-12
  Natural Language Processing

  Compressing word embeddings with low-rank matrix factorization for on-device natural language understanding.

  Anish Acharya, A Metallinou, R Goel, I Dhillon

  US10872601 · Amazon · USPTO

Papers

2026

• conferenceSPLICE: Structured Prompt Local Iterative Combinatorial Evolution

  Anish Acharya, P Studans, A Dhanda, N V Rao, V A Khatri, S A Niaki, B Verkhovsky

  Conference on Neural Information Processing Systems (NeurIPS), 2026

  Treats a production prompt as a set of blocks, some frozen and some editable, and optimizes it with section-local textual gradients and elitist beam search, giving the first formal guarantees for iterative LLM-driven prompt optimization. Outperforms prior prompt optimizers across benchmarks and three model capability tiers while keeping frozen sections verbatim.

• workshopRoPoLL: Robust Panel of LLM Judges

  Anish Acharya, K Pan, B Verkhovsky

  International Conference on Machine Learning (ICML), 2026 · AgenticUQ Workshop

  Shows that averaging an LLM jury is unboundedly biased once any judge fails systematically, and replaces the average with a geometric-median aggregator that has the optimal 1/2 breakdown point. A 3-judge 38B panel beats a 675B judge by 1.31x under 30% corruption.

  OpenReviewarXivAmazon Science

• workshopOrchOpt: Validation-Gated Architecture Search for Self-Organizing Multi-Agent Orchestration

  A Dhanda, S A Niaki, Anish Acharya, B Verkhovsky

  Conference on Language Modeling (COLM), 2026 · Lifelong Agents Workshop

  Searches over multi-agent orchestration architectures and keeps a change only when it passes validation.

  OpenReviewAmazon Science

2025

• conferenceGeometric Median Matching for Robust k-Subset Selection from Noisy Data

  Anish Acharya, S Sanghavi, A G Dimakis, I S Dhillon

  International Conference on Machine Learning (ICML), 2025

  Selects a k-subset whose mean tracks the geometric median of noisy data, giving an O(1/k) rate under arbitrary corruption, a quadratic improvement over random sampling.

  PMLROpenReviewarXiv

  Short version: ICML 2024 FMWild Workshop

2024

• journalNeural Distributed Source Coding

  A Nagle, J Whang, Anish Acharya, H Kim, A Dimakis

  IEEE Journal on Selected Areas in Information Theory (IEEE JSAIT), 2024

  Learns lossy distributed source coding with a conditional VQ-VAE, handling arbitrary correlation structure in high dimensions with state-of-the-art PSNR.

  IEEE XplorearXiv

2022

• conferenceFaster Non-Convex Federated Learning via Global and Local Momentum

  R Das, Anish Acharya, A Hashemi, S Sanghavi, I Dhillon, U Topcu

  Conference on Uncertainty in Artificial Intelligence (UAI), 2022 · Spotlight

  Combines variance-reduced momentum at the server and at the clients to speed up non-convex federated learning under client heterogeneity and compressed communication.

  OpenReviewarXiv

• conferenceRobust Training in High Dimensions via Block Coordinate Geometric Median Descent

  Anish Acharya, A Hashemi, P Jain, S Sanghavi, I Dhillon, U Topcu

  International Conference on Artificial Intelligence and Statistics (AISTATS), 2022 · Spotlight

  Makes geometric-median SGD practical in high dimensions by applying it to one chosen block of coordinates at a time, keeping the optimal 1/2 breakdown point for smooth non-convex problems.

  PMLRCodePosterSlides

  Short version: Joint IFML/CCSI Symposium (Simons Institute, UC Berkeley) · Short version: NSF-TRIPODS Workshop on Communication Efficient Distributed Optimization

• workshopPositive Unlabeled Contrastive Learning

  Anish Acharya, S Sanghavi, L Jing, B Bhushanam, D Choudhary, M Rabbat, I Dhillon

  International Conference on Machine Learning (ICML), 2022 · PODS Workshop

  Extends contrastive pretraining to positive-unlabeled data with an unbiased, lower-variance loss that uses the few labeled positives.

  ProceedingsarXiv

  Extended version: Understanding Contrastive Representation Learning from Positive Unlabeled (PU) Data — arXiv:2402.06038 · Short version: Simons Institute, UC Berkeley — Data-Driven Decision Processes Workshop

• workshopLDKP: A Dataset for Identifying Keyphrases from Long Scientific Documents

  D Mahata, N Agarwal, D Gautam, A Kumar, S Parekh, Y K Singla, Anish Acharya, R R Shah

  ACM International Conference on Information and Knowledge Management (CIKM), 2022 · DL4SR Workshop

  Releases two corpora, about 1.3M and 100K full-text scientific articles, for identifying keyphrases beyond the title and abstract.

  CEUR-WSarXivHF Dataset

2021

• journalOn the Benefits of Multiple Gossip Steps in Communication Constrained Federated Learning

  A Hashemi, Anish Acharya, R Das, H Vikalo, S Sanghavi, I Dhillon

  IEEE Transactions on Parallel and Distributed Systems (IEEE TPDS), 2021

  Shows that several gossip steps between gradient updates improve decentralized learning under lossy compressed communication, even after paying for the extra communication.

  IEEE XploreCodearXiv

• conferenceAlexa Conversations: An Extensible Data-driven Approach for Building Task-oriented Dialogue Systems

  Anish Acharya, S Adhikari, S Agarwal, V Auvray, N Belgamwar, A Biswas, S Chandra, T Chung, M Zarandi, R Gabriel, S Gao, R Goel, D Hakkani-Tur, J Jezabek, A Jha, J Kao, P Krishnan, P Ku, A Goyal, C Lin, Q Liu, A Mandal, A Metallinou, V Naik, Y Pan, S Paul, V Perera, A Sethi, M Shen, N Strom, E Wang

  Conference of the North American Chapter of the Association for Computational Linguistics (NAACL), 2021 · Demonstrations Track

  Builds task-oriented dialogue systems from a few seed dialogues and API specifications by simulating training dialogues. The simulator improves turn-level action prediction accuracy by over 50%.

  ACL AnthologySkill

• conferenceGupShup: An Annotated Corpus for Abstractive Summarization of Open-Domain Code-Switched Conversations

  L Mehnaz, D Mahata, A Kumar, U S Gunturi, R Jain, G Gupta, R Gosangi, I G Lee, Anish Acharya, R Shah

  Conference on Empirical Methods in Natural Language Processing (EMNLP), 2021

  Introduces the first summarization dataset of Hindi-English code-switched conversations: over 6,800 conversations with human-written summaries.

  ACL AnthologyCode

2019

• conferenceOnline Embedding Compression for Text Classification using Low Rank Matrix Factorization

  Anish Acharya, R Goel, A Metallinou, I Dhillon

  AAAI Conference on Artificial Intelligence (AAAI), 2019 · Spotlight

  Compresses the word-embedding layer during training with low-rank factorization, reaching 90% compression with minimal accuracy loss and no added latency.

  ProceedingsCodearXiv

2014

• journalExtending The Concept of Analog Butterworth Filter For Fractional Domain

  Anish Acharya, S Das, I Pan, S Das

  Signal Processing (Elsevier), 2014

  Designs fractional-order Butterworth filters in the complex w-plane, including under-, hyper-, and ultra-damped poles, and demonstrates the formulation in simulations and a frequency-domain design example.

  ScienceDirectarXiv

• journalSimulation studies on the design of optimum PID controllers to suppress chaotic oscillations in a family of Lorenz-like multi-wing attractors

  S Das, Anish Acharya, I Pan

  Mathematics and Computers in Simulation (Elsevier), 2014

  Uses real-coded genetic-algorithm optimization to tune PID controllers for four Lorenz-family multi-wing chaotic systems, with robustness checked across initial conditions.

  ScienceDirectarXiv

2013

• conferenceStability Analysis Of Delayed System Using Bode's Integral

  Anish Acharya, D Mitra, K Halder

  International Conference on Computer Communication and Informatics (ICCCI), 2013

  Tunes PID controllers for delayed plants using Bode's integral and a Padé approximation of the delay, then compares the approach with genetic-algorithm tuning in MATLAB simulations.

  IEEE XplorearXiv

2012

• conferenceOptimum PID Control of Multi-wing Attractors in A Family of Lorenz-like Chaotic Systems

  Anish Acharya, S Das, I Pan

  International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2012

  Applies real-coded genetic-algorithm optimization to tune PID control of Lorenz-like multi-wing chaotic attractors and compares the resulting state trajectories.

  IEEE XplorearXiv

• conferenceOptimized Quality Factor of Fractional Order Analog Filters with Band-Pass and Band-Stop Characteristics

  A Pakhira, S Das, Anish Acharya, I Pan, S Saha

  International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2012

  Studies fractional-order band-pass and band-stop filters, whose characteristics are unavailable to conventional integer-order filters below second order, and optimizes their quality factor.

  IEEE XplorearXiv

• conferenceIdentification of nonlinear systems from the knowledge around different operating conditions: a feed-forward multilayer Artificial Neural Net (ANN) based approach

  S Saha, S Das, Anish Acharya, A Kumar, S Mukherjee, I Pan, A Gupta

  International Conference on Parallel, Distributed and Grid Computing (PDGC), 2012

  Identifies nonlinear systems from outputs near several linearized operating points using feed-forward multilayer neural networks, evaluated on nuclear-reactor monitoring and AC-servo control.

  IEEE Xplore

2011

• conferenceLeast square and instrumental variable system identification of AC servo position control system with fractional Gaussian noise

  S Das, A Kumar, I Pan, Anish Acharya, S Das, A Gupta

  International Conference on Energy, Automation and Signal (ICEAS), 2011

  Compares least-squares and instrumental-variable estimators for identifying an AC-servo position-control system under white and fractional Gaussian measurement noise.

  IEEE Xplore