{"id":"a62059d8-9c28-465a-8d80-8290479b9430","arxiv_id":"2505.19996","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"OMIB introduces a bound on the MIB redundancy weight and per-modality adaptive weighting, but the proof of the bound uses a different objective and unjustified independence.","lead":"OMIB is a multimodal representation learning framework that claims to achieve an optimal information bottleneck by setting a regularization weight within a data-dependent bound and adapting weights per modality. If the theory held, it would offer a principled alternative to hand-tuned MIB regularization; the paper reports gains on sentiment, emotion, and tissue detection benchmarks but the proofs have gaps.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Prop. 5.1's variational bound drops I(ζ2;z1|ζ1) via a false independence: ζ1 and ζ2 are correlated through shared a0/b0, so L_OMF is not an upper bound for Eq. (17), breaking the theory-to-objective link in the central guarantee.","rationale":"The reader's REJECT is justified. The strongest formal claim is Proposition 5.7, but it concerns Eq. (17), while the objective actually trained is L_OMF (Eq. (10)). The only argument connecting them is Proposition 5.1's variational upper bound. That proof is invalid: the omitted conditional mutual information is not zero under the paper's own assumption that modalities share a0/b0. Because z1 and z2 are correlated, ζ2 provides additional information about z1 beyond ζ1, so Eq. (40) is false. Consequently, the learned objective may impose a different trade-off than Eq. (17), and the beta bound computed from entropies of v1,v2 does not guarantee F(ξ)={a0,a1,a2} for the actual loss. The other reader-identified issue, replacing I(ξ;y) with relative mutual information in Appendix C, is real but secondary: re-deriving Lemmas 5.5 and 5.6 directly with I(ξ;y) shows the inclusion/exclusion comparisons still hold under beta < Mu, so that flaw may be repairable. The Proposition 5.1 flaw is not repairable without adding cross terms such as I(ζ2;z1|ζ1) plus I(ζ1;z2|ζ2) to L_OMF, or without assuming independent modalities, which contradicts Assumption 5.3. Therefore the theory-to-algorithm bridge fails, and the paper's headline guarantee is unsupported. The empirical results may still support the heuristic method, but the central theoretical claim is not established, so the verdict remains REJECT.","tokens_in":38933,"tokens_out":13076,"duration_ms":128551,"concrete_test":"Run OMIB on SIM-I (shared dimensions d0=d0'=200) after warm-up, freeze the encoders, draw 10k paired samples, and estimate the omitted conditional mutual information I(ζ2;z1|ζ1) with a k-NN estimator or MINE. If the estimate is significantly above zero, Eq. (40) is false and the loss in Eq. (10) is not a variational upper bound on Eq. (17), so the claimed beta-guarantee for the trained objective does not follow.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Appendix B, Eq. (40) proves I(ξ;z1) ≤ I(ζ1,ζ2;z1) = I(ζ1;z1) + I(ζ2;z1|ζ1), then sets I(ζ2;z1|ζ1)=0, citing the 'Markov property' ζ1⊥z2 and ζ2⊥z1. This independence is not implied by the generative model. ζ1 is conditionally independent of z2 given z1, but not marginally independent of z2, and z1,z2 are correlated by construction under Assumption 5.3 because they share a0 and b0. After the VAE reparameterization, ζ2 is a noisy function of the shared component, so I(ζ2;z1|ζ1) > 0 whenever modalities share information. Dropping this term makes L_OMF strictly smaller than the claimed upper bound; the regularizer underestimates I(ξ;z1), so the beta interval (0,Mu] derived for Eq. (17) no longer controls the objective actually minimized in Eq. (10). Section 4 explicitly presents L_OMF as the loss that achieves optimal MIB via Proposition 5.1; without that bridge, Proposition 5.7 does not apply to the trained model. The Appendix C switch from I to relative mutual information is a separate proof gap, but the failure of Proposition 5.1 is sufficient to break the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes OMIB, a multimodal information bottleneck framework whose optimization objective allegedly guarantees the achievability of an optimal multimodal information bottleneck. The central theoretical claim is Proposition 5.7: under Assumption 5.3, optimizing the objective in Eq. (17) with a regularization weight β in (0, Mu], where Mu is derived from the entropies and mutual information of the input modalities, yields a representation ξ whose information set is exactly {a0, a1, a2}. The method adds a dynamically computed per-modality weight r and implements the objective via variational approximations (Eq. (10)). The paper reports experiments on synthetic data and several real-world multimodal benchmarks.","tokens_in":39189,"tokens_out":7426,"duration_ms":74924,"significance":"If the central guarantee were correct, the contribution would be valuable: it would replace ad hoc selection of IB regularization weights with a computable a priori bound and would explicitly address imbalance in task-relevant information across modalities. The synthetic experiment in Section 6.2 directly tests the β bound and shows accuracy peaking near Mu, which is a good falsifiable check, and the empirical comparisons are extensive. However, the theoretical bridge from Eq. (17) to the implemented loss (10) and to the proofs of Lemmas 5.5 and 5.6 contains load-bearing gaps, so the main guarantee is not established as written.","major_comments":[{"comment":"The derivation of the upper bound I(ξ; z1) ≤ I(ζ1; z1) rests on the asserted Markov property ζ2 ⊥ z1, but this independence is not implied by the generative model: under Assumption 5.3, v1 and v2 share a0 and b0, and after the reparameterization in Eq. (6) both ζ1 and ζ2 are noisy functions of these shared components, so I(ζ2; z1|ζ1) > 0 in general. Dropping this term makes L_OMF in Eq. (10) strictly smaller than the claimed variational upper bound on Eq. (17), so the β interval (0, Mu] derived for Eq. (17) cannot be applied to the objective that is actually minimized during training. This breaks the theory-to-objective link on which Proposition 5.7 relies.","section":"Appendix B, Eq. (40)"},{"comment":"The proofs of Lemmas 5.5 and 5.6 explicitly replace I(ξ; y) with the relative mutual information Î(ξ; y) \"in the following analysis,\" but no argument is given that the two objectives share the same optima or that the same β bound Mu applies to the modified objective. Proposition 5.7 is stated for Eq. (17), which uses I(ξ; y), while the proof establishes properties of a different objective. This is a gap at the core of the achievability claim and is acknowledged in the manuscript's own text.","section":"Appendix C, Definition C.1 and Lemmas C.2/C.4"},{"comment":"The dynamic weight r is computed from KL divergences between the model's own predictive distributions p(ŷ_i|ξ,z_i) and p(ŷ|ξ). Since these same distributions are being optimized in Eq. (10), the proof that Eq. (11) satisfies the proportionality in Eq. (18) assumes access to the true posterior p(y|·), which the model only approximates. The derived r is therefore self-referential; Proposition 5.2 does not establish the claimed behavior for the quantity actually evaluated during training.","section":"Section 5.1, Proposition 5.2 and Eq. (11)"},{"comment":"The variational loss in Eq. (10) regularizes the KL divergence of ζ_i with N(0,I), but the objective in Eq. (17) is written in terms of I(ξ; z_i). The injected noise ϵ in the reparameterization (6) can propagate to ξ without increasing either I(ξ; z1) or I(ξ; z2), so the exclusiveness proof in Lemma 5.6, which only considers superfluous information contained in v1 or v2, does not rule out noise-only components in ξ. Hence the claimed equality F(ξ) = {a0,a1,a2} is not enforced by the optimized loss.","section":"Appendix B, Eqs. (42)–(45), and Lemma 5.6"}],"minor_comments":[{"comment":"The symbol Mu is used both in Lemma 5.5 as 1/((1+r)(H(v1)+H(v2)-I(v1;v2))) and then redefined in Section 5.2 as 1/(3(H(v1)+H(v2)-I(v1;v2))) without a distinct name; this will confuse readers about which bound Proposition 5.7 refers to.","section":"Section 5.2"},{"comment":"There are typos: 'Langrange' should be 'Lagrange', and 'tahn' should be 'tanh'.","section":"Section 5.1"},{"comment":"The caption lists case iii as having F_rel(v1)=F_rel(v2) with d11(300)=d21(300), but the text in Section 6.2 says the balanced dataset is SIM-III; please make the correspondence between the three cases and the SIM datasets explicit in the caption.","section":"Figure 3"},{"comment":"The KL terms in Eq. (10) are written without a batch average; if they are averaged over the batch as the empirical expectation in the preceding line suggests, the notation should be consistent.","section":"Equation (10)"}],"recommendation":"reject","confidential_remarks":"The manuscript has a substantial empirical section and a clear presentation, but the central theoretical guarantee is invalid as written: the variational bound in Proposition 5.1 relies on an unjustified independence assumption, and the proofs of Lemmas 5.5 and 5.6 change the objective being optimized. These are not local presentation issues; repairing them would require a new derivation of the β bound and a demonstration that the training objective is actually an upper bound on the theoretical objective. As submitted, the paper does not establish its core claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper proposes OMIB, a multimodal information bottleneck method that adapts the per-modality regularization weight and derives a data-dependent upper bound on the redundancy coefficient β. The formulation is sensible: rather than tuning β and the modality weights by hand, compute them from the data. The empirical results on CREMA-D, MOSI, and the tissue-detection benchmarks are competitive, and the synthetic experiment showing a peak in accuracy near the proposed β bound is interesting.\n\nWhat's actually new is the dynamic weight r (Eq. 11) that grows when a modality still has task-relevant information missing from the fused representation, and the explicit Mu bound. I don't see those in the cited MIB papers. The paper also tries to connect the objective to the five information-theoretic factors, which is a useful framing.\n\nThe soft spots, however, are load-bearing. The proof of Proposition 5.1 in Appendix B assumes a 'Markov property' ζ1⊥z2 and ζ2⊥z1, and drops the term I(ζ2; z1|ζ1). That independence is not implied by the generative model. Since v1 and v2 share a0 and b0, the encodings z1 and z2 are correlated, and the reparameterized ζ1, ζ2 inherit that correlation. The dropped term is positive, so L_OMF is a lower bound, not an upper bound, on the objective in Eq. (17). The β interval (0, Mu] derived for that objective therefore doesn't control what's actually minimized. The paper's central claim rests on this bridge, and it's broken.\n\nSeparately, the proofs of Lemmas 5.5 and 5.6 in Appendix C replace I(ξ;y) with the relative mutual information Ĩ(ξ;y) and then analyze that modified objective. They never show that the two objectives share the same optimizer. So the achievability theorem is about a different loss than the one trained. The set-algebra treatment of information atoms is also informal; it would need to be made rigorous.\n\nThe empirical reporting has issues. In Table 3, the best non-MIB method is BiGated (58.4), not concatenation (53.2), so the claimed improvement margin is wrong. In Table 4, OMIB ties E-MIB on Acc7, yet the text says it outperforms all baselines on all metrics. No error bars are given, and the mean tissue-detection numbers mask large per-dataset variation.\n\nThat said, the heuristic itself may have value. The dynamic r is a reasonable response to modality imbalance, and the gains on CREMA-D and MOSI are real. The synthetic validation is a nice idea, but it doesn't rescue the proof.\n\nBottom line: a sensible engineering idea with a broken proof of its central guarantee. The REJECT verdict is correct. I'd still send it to peer review, because the flaws are substantive and a careful referee could identify them clearly, possibly pointing the authors toward a fix or a weaker claim. It's also a good case study for a reading group on how a false independence assumption can invalidate a variational bound.","headline":"A sensible heuristic with a broken proof: the claimed guarantee on optimal MIB fails because the variational bound drops a positive term and the appendix changes the objective.","tokens_in":39817,"tokens_out":4517,"would_cite":false,"duration_ms":45575,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper argues that a single regularization weight, chosen inside a provable interval, makes a multimodal bottleneck representation achieve the theoretical optimum: it keeps all task-relevant information and nothing else.","keywords":["multimodal learning","information bottleneck","multimodal information bottleneck","regularization weight bound","variational inference","mutual information estimation","modality imbalance","representation learning"],"falsifier":"On the paper's own synthetic Gaussian datasets (SIM-I/II/III), after training compute the dropped conditional-information term $I(\\zeta_2; z_1 | \\zeta_1)$: wherever the modalities share the blocks $a_0$ and $b_0$, this term is positive, and if it is positive the inequality $I(\\xi;z_1) \\le I(\\zeta_1;z_1)$ used in Proposition 5.1 is not established. A second check targets the relative-information substitution directly: train with a large shared task-relevant block $a_0$ and $\\beta$ just below $M_u$, then measure whether the learned $\\xi$ retains all of $a_0$ and none of $b_0$; a regime where the relative and ordinary objectives disagree would show a representation that omits part of $a_0$ or keeps part of $b_0$, contradicting Proposition 5.7.","tokens_in":38595,"feed_emoji":"🎯","tokens_out":19623,"duration_ms":160491,"temperature":0.7,"pith_summary":"Multimodal information bottleneck (MIB) methods try to learn a fused representation that keeps what is needed to predict a task and discards the rest, but they tune the strength of the 'discard' term by hand and can drop a weak modality's crucial information. This paper tries to close that gap by proving that a single regularization weight $\\beta$, set inside the bound $M_u = 1/((1+r)(H(v_1)+H(v_2)-I(v_1;v_2)))$, forces the fused representation to contain exactly the task-relevant information $\\{a_0, a_1, a_2\\}$ and no superfluous information $\\{b_0, b_1, b_2\\}$. The framework it proposes, OMIB, pairs that objective with a per-modality weight $r$ that shrinks automatically when one modality still holds undigested task information, so imbalanced modalities are not regularized out. If the proof holds, the bottleneck strength stops being an ad hoc knob: both ends of the valid range are computable from the training data before training, via mutual information estimates. The paper validates the bound on controlled Gaussian data, where accuracy peaks just below $M_u$, and reports downstream gains on emotion recognition, multimodal sentiment analysis, and anomalous tissue detection.","feed_headline":"One bound provably yields the optimal multimodal bottleneck","feed_subtitle":"Pick the strength from a data-derived range; the representation then keeps exactly the task-relevant information.","key_machinery":"The load-bearing object is the objective of Equation (17), $\\min_\\xi \\ell(\\xi) = -I(\\xi;y) + \\beta(I(\\xi;z_1) + r I(\\xi;z_2))$, together with the bound $M_u = 1/((1+r)(H(v_1)+H(v_2)-I(v_1;v_2)))$ on the redundancy weight $\\beta$. Assumption 5.3 supplies the Venn-diagram decomposition of the two modalities into task-relevant atoms $a_0, a_1, a_2$ and superfluous atoms $b_0, b_1, b_2$; this is what lets the proof treat missing task information as an atom disjoint from the current representation and show that adding it strictly lowers the loss whenever $\\beta < M_u$. A second device carries the proof: the derivations of Lemmas 5.5 and 5.6 replace $I(\\xi;y)$ with the relative mutual information $\\hat{I}(\\xi;y) = I(\\xi;y)/(H(\\xi)+H(y)-I(\\xi;y))$, arguing that this normalized quantity better tracks the share of task-relevant information in the total. The objective actually trained is the variational upper bound of Equation (10), which converts $I(\\xi;y)$ into a cross-entropy term and each $I(\\xi;z_i)$ into a KL divergence against a standard Gaussian.","core_discovery":"On the paper's own terms, the discovery is that the optimal multimodal bottleneck is achievable, not merely approachable. Under Assumption 5.3, which partitions each modality's information into task-relevant atoms (the shared $a_0$ and the modality-specific $a_1, a_2$) and superfluous atoms ($b_0$ shared, $b_1, b_2$ specific), the optimal MIB is defined by $F(\\xi_{\\mathrm{opt}}) = \\{a_0, a_1, a_2\\}$. The paper proves that minimizing $\\ell(\\xi) = -I(\\xi;y) + \\beta(I(\\xi;z_1) + r I(\\xi;z_2))$ with $\\beta \\in (0, M_u]$ both pulls all task-relevant information in (Lemma 5.5) and pushes all superfluous information out (Lemma 5.6), so that $F(\\xi) = \\{a_0, a_1, a_2\\}$ (Proposition 5.7). The dynamic weight $r$ is computed during training from the ratio of the two modalities' remaining task-relevant information, which keeps the weaker modality's complementary information (say $a_2$) from being penalized away. The whole objective is implemented as a variational upper bound, so the intractable mutual informations become a cross-entropy term plus KL divergences.","pith_inferences":["The atomized Venn-diagram model ($a_0/a_1/a_2$ versus $b_0/b_1/b_2$) is an idealization; for real modalities with graded overlaps the practical content of Proposition 5.7 is a regime statement, and a testable extension is to intervene on the $a_0$ and $b_0$ blocks of the synthetic data to map how quickly the guarantee degrades as the shared superfluous block $b_0$ grows.","The proof's unstated swap to relative mutual information suggests a cleaner route: if the objective were written directly in terms of the normalized measure, the two-objective gap would disappear, and one could check empirically whether the OMIB bound still holds under that rewrite.","The bound $M_u$ is derived from $F(\\xi,y) \\subseteq F(v_1,v_2)$; computing the slack $F(v_1,v_2) - F(\\xi,y)$ during training would give an online measure of how far the current representation is from the claimed optimum, effectively a certificate that the representation has not yet converged.","If the conditional-independence violation cannot be patched, a natural repair is to add the dropped term $I(\\zeta_2;z_1|\\zeta_1)$ as an explicit penalty, and to test whether the interval for $\\beta$ widens or narrows as a result."],"forward_implications":["Bottleneck strength stops being a free knob: with $M_l$ and $M_u$ estimated from the training data by a neural mutual-information estimator, a practitioner can set $\\beta$ in the guaranteed range before training and get the optimal representation without per-task search.","Imbalanced modalities are protected: because $r$ shrinks when one modality still holds undigested task-relevant information, a modality with minor but crucial information (such as $a_2$ in the paper's Figure 1) is not regularized away, so the fused representation stays sufficient.","The guarantee extends to three or more modalities through the generalized bound, so the argument covers fusion beyond the pairwise case treated in the main text.","Because the optimizable loss is a variational upper bound, the theoretical optimum is approached by minimizing a standard cross-entropy plus KL-divergence objective, so the recipe drops into existing deep learning pipelines.","On the paper's reported tasks, following the bound produces representations that outperform fixed-weight MIB baselines on emotion recognition, sentiment analysis, and anomalous tissue detection."],"supporting_citations":[{"why":"Supplies the variational upper-bound construction that turns the intractable mutual informations in the OMIB objective into the cross-entropy and KL-divergence terms of the trainable loss.","marker":"(Alemi et al., 2017)"},{"why":"Introduces the deep multi-view IB objective $\\max_z I(y;z) - \\beta \\sum_i I(v_i;z)$ that Equation (1) generalizes and that OMIB's objective is built on.","marker":"(Wang et al., 2019)"},{"why":"Formulates the multi-view information bottleneck with sufficiency objectives, motivating the task-relevance branches and the two-part IB principle in Equation (12).","marker":"(Federici et al., 2020)"},{"why":"Provides the neural mutual-information estimator (MINE) used to compute $H(v_i)$ and $I(v_1;v_2)$, which makes $M_l$ and $M_u$ computable a priori from training data.","marker":"(Belghazi et al., 2018)"},{"why":"Supplies the entropy and mutual-information identities (Properties A.1 in Appendix A) that the proofs of Lemmas 5.5 and 5.6 rely on.","marker":"(Cover, 1999)"},{"why":"Defines the original information bottleneck principle of maximizing label-relevant information while compressing the input, which the OMIB objective formalizes.","marker":"(Tishby et al., 2000)"},{"why":"Provides the E-MIB, L-MIB, and C-MIB baselines and the CMU-MOSI experimental setup against which OMIB is evaluated.","marker":"(Mai et al., 2023)"}],"fun_headline_variants":["Optimal multimodal bottleneck is now provably achievable","Per-modality weights provably reach the optimal bottleneck","One derived bound yields the exact optimal multimodal bottleneck","Provable bound guarantees optimal multimodal bottleneck"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The guarantee rests on assuming the two modality encodings are conditionally independent, $\\zeta_1 \\perp z_2$ and $\\zeta_2 \\perp z_1$, when the variational bound is derived; the paper's own setup violates this because the modalities share the $a_0$ and $b_0$ blocks, and the proofs of the two lemmas also swap in the relative mutual information without showing that it shares the original objective's optimum.","fun_headline_variants_meta":{"raw":{"variants":["Optimal multimodal bottleneck is now provably achievable","Per-modality weights provably reach the optimal bottleneck","One derived bound yields the exact optimal multimodal bottleneck","Provable bound guarantees optimal multimodal bottleneck"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000942,"raw_usage":{"total_tokens":4062,"prompt_tokens":1022,"completion_tokens":3040,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":638,"completion_tokens_details":{"reasoning_tokens":2982}},"tokens_in":638,"tokens_out":3040,"duration_ms":23921,"temperature":1.0,"reasoning_tokens":2982,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:02:37.335903+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On the paper's own synthetic Gaussian datasets (SIM-I/II/III), after training compute the dropped conditional-information term $I(\\zeta_2; z_1 | \\zeta_1)$: wherever the modalities share the blocks $a_0$ and $b_0$, this term is positive, and if it is positive the inequality $I(\\xi;z_1) \\le I(\\zeta_1;z_1)$ used in Proposition 5.1 is not established. A second check targets the relative-information substitution directly: train with a large shared task-relevant block $a_0$ and $\\beta$ just below $M_u$, then measure whether the learned $\\xi$ retains all of $a_0$ and none of $b_0$; a regime where the relative and ordinary objectives disagree would show a representation that omits part of $a_0$ or keeps part of $b_0$, contradicting Proposition 5.7.","supporting_citations":[{"cited_title":"Deep multi-view information bottleneck","cited_arxiv_id":null,"evidence_quote":"Introduces the deep multi-view IB objective $\\max_z I(y;z) - \\beta \\sum_i I(v_i;z)$ that Equation (1) generalizes and that OMIB's objective is built on."}],"review_version":1}