Aditya Ahuja

Jean Erik Delanois, Aditya Ahuja, Giri P. Krishnan et al.

Artificial neural networks are often overconfident, undermining trust because their predicted probabilities do not match actual accuracy. Inspired by biological sleep and the role of spontaneous replay in memory and learning, we introduce Sleep Replay Consolidation (SRC), a novel calibration approach. SRC is a post-training, sleep-like phase that selectively replays internal representations to update network weights and improve calibration without supervised retraining. Across multiple experiments, SRC is competitive with and complementary to standard approaches such as temperature scaling. Combining SRC with temperature scaling achieves the best Brier score and entropy trade-offs for AlexNet and VGG19. These results show that SRC provides a fundamentally novel approach to improving neural network calibration. SRC-based calibration offers a practical path toward more trustworthy confidence estimates and narrows the gap between human-like uncertainty handling and modern deep networks.

Aditya Ahuja, Aditya Lahiri, Aniruddha Das

Figuring out the price of a listed Airbnb rental is an important and difficult task for both the host and the customer. For the former, it can enable them to set a reasonable price without compromising on their profits. For the customer, it helps understand the key drivers for price and also provides them with similarly priced places. This price prediction regression task can also have multiple downstream uses, such as in recommendation of similar rentals based on price. We propose to use geolocation, temporal, visual and natural language features to create a reliable and accurate price prediction algorithm.

Tirtharaj Dash, Sharad Chitlangia, Aditya Ahuja et al.

We present a survey of ways in which existing scientific knowledge are included when constructing models with neural networks. The inclusion of domain-knowledge is of special interest not just to constructing scientific assistants, but also, many other areas that involve understanding data using human-machine collaboration. In many such instances, machine-based model construction may benefit significantly from being provided with human-knowledge of the domain encoded in a sufficiently precise form. This paper examines the inclusion of domain-knowledge by means of changes to: the input, the loss-function, and the architecture of deep networks. The categorisation is for ease of exposition: in practice we expect a combination of such changes will be employed. In each category, we describe techniques that have been shown to yield significant changes in the performance of deep neural networks.

Aditya Ahuja, Vinay J. Ribeiro, Ranjan Pal

Permissionless blockchain consensus protocols have been designed primarily for defining decentralized economies for the commercial trade of assets, both virtual and physical, using cryptocurrencies. In most instances, the assets being traded are regulated, which mandates that the legal right to their trade and their trade value are determined by the governmental regulator of the jurisdiction in which the trade occurs. Unfortunately, existing blockchains do not formally recognise proposal of legal cryptocurrency transactions, as part of the execution of their respective consensus protocols, resulting in rampant illegal activities in the associated crypto-economies. In this contribution, we motivate the need for regulated blockchain consensus protocols with a case study of the illegal, cryptocurrency based, Silk Road darknet market. We present a novel regulatory framework for blockchain protocols, for ensuring legal transaction confirmation as part of the blockchain distributed consensus. As per our regulatory framework, we derive conditions under which legal transaction throughput supersedes throughput of traditional transactions, which are, in the worst case, an indifferentiable mix of legal and illegal transactions. Finally, we show that with a small change to the standard blockchain consensus execution policy (appropriately introduced through regulation), the legal transaction throughput in the blockchain network can be maximized.

Tirtharaj Dash, Sharad Chitlangia, Aditya Ahuja et al.

We present a survey of ways in which domain-knowledge has been included when constructing models with neural networks. The inclusion of domain-knowledge is of special interest not just to constructing scientific assistants, but also, many other areas that involve understanding data using human-machine collaboration. In many such instances, machine-based model construction may benefit significantly from being provided with human-knowledge of the domain encoded in a sufficiently precise form. This paper examines two broad approaches to encode such knowledge--as logical and numerical constraints--and describes techniques and results obtained in several sub-categories under each of these approaches.

Aditya Ahuja

Strategic interactions between competitive entities are generally considered from the perspective of complete revelation of benefits achieved from those interactions, in the form of public payoff functions and/or beliefs, in the announced games. However, there exist strategic interplays between competitors where the players have a choice to strategise under the availability of private payoffs, in similar competitive settings. In this contribution, we propose a formal framework for a competitive ecosystem where each player is permitted to defect from publicly optimal strategies under certain private payoffs greater than announced payoffs, given that these defections have certain acceptable bounds in the long run as agreed by all players. We call this game theoretic construction an Intention Game. We formally define an Intention Game, and notions of participational equilibria that exist in such interactions that permit public defections. We compare Intention Games with conventional strategic form games, and demonstrate a type-theoretic construction of Intention Games. In a partially honest setting, we give Intention Game instances of a Cournot competition, secure interactions between mobile applications, an Internet services' data sourcing competition between Internet service providers through content delivery networks, and a Bitcoin mining competition. We give a use of Intention Games to determine player participation in a cryptographic protocol. Finally, we demonstrate the possibility of a dual model of the Intention Games framework.

Aditya Ahuja

Quantum encryption is a well studied problem for both classical and quantum information. However, little is known about quantum encryption schemes which enable the user, under different keys, to learn different functions of the plaintext, given the ciphertext. In this paper, we give a novel one-bit secret-key quantum encryption scheme, a classical extension of which allows different key holders to learn different length subsequences of the plaintext from the ciphertext. We prove our quantum-classical scheme secure under the notions of quantum semantic security, quantum entropic indistinguishability, and recent security definitions from the field of functional encryption.